Image screening method, identification method, device and electronic equipment

Abstract

Embodiments of the present application provide an image screening method, an identification method, an apparatus and an electronic device, and relate to the technical field of machine vision. The image screening method comprises: obtaining an identification result obtained by performing image identification processing on at least one candidate image; obtaining at least one image feature of the candidate image; wherein the image feature is a feature capable of representing the imaging quality of the to-be-identified content; and determining an image screening result from the at least one candidate image based on the identification result of the candidate image and the at least one image feature. Through the image screening method, the accuracy of image screening can be improved.

Description

Technical Field

[0001] This application relates to the field of machine vision technology, and in particular to an image screening method, recognition method, device and electronic device. Background Technology

[0002] For image tasks such as image recommendation, image filtering is usually performed based on multiple candidate images (such as multiple images of the same target object containing character content taken with different imaging configuration information). The task result is determined based on the selected candidate images (e.g., determining the imaging recommendation result containing imaging configuration information). The image filtering is used to select images with higher imaging quality (such as images with clear character imaging).

[0003] In related technologies, image recognition processing (such as text recognition) is performed on each of the multiple candidate images, and the recognition results are used as the basis for image selection.

[0004] However, the relevant technologies rely solely on the recognition results as the basis for selection, making image selection entirely dependent on the recognition effect of image recognition processing. This results in a low accuracy rate for image selection, which ultimately affects the accuracy of image recommendation. Summary of the Invention

[0005] The purpose of this application is to provide an image filtering method, recognition method, apparatus, and electronic device. The specific technical solution is as follows:

[0006] In a first aspect, embodiments of this application provide an image filtering method, including:

[0007] Obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0008] Obtain at least one image feature of the candidate image; wherein the image feature is a feature capable of characterizing the imaging quality of the content to be identified;

[0009] Based on the recognition results of the candidate images and the at least one image feature, an image filtering result is determined from the at least one candidate image.

[0010] Secondly, embodiments of this application also provide an image filtering method, including:

[0011] Obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0012] Image recognition processing is performed on the reference image of the candidate image to obtain the reference content of the candidate image; wherein the content to be identified in the reference image of the candidate image is the same as the content to be identified in the candidate image.

[0013] Based on the reference content of the candidate image, the target auxiliary evaluation index of the candidate image is determined;

[0014] Based on the recognition results of the candidate images and the target auxiliary evaluation index, the image screening results are determined from the at least one candidate image.

[0015] Thirdly, embodiments of this application provide an identification method applied to a client, the method comprising:

[0016] Receive configuration requests for the sample;

[0017] In response to the configuration request, a first recognition result and a first evaluation index for the sample are displayed; the first evaluation index indicates the imaging quality of the corresponding first recognition result; the first recognition result is the recognition result obtained by performing image recognition processing on multiple imaging results of the sample, and the corresponding first evaluation index satisfies the first screening condition; the multiple imaging results are obtained by imaging the sample based on set multi-group imaging conditions.

[0018] In response to user interaction, the imaging conditions corresponding to the first recognition result are set to the first imaging conditions for the sample to complete the configuration.

[0019] Fourthly, embodiments of this application provide an image filtering device, comprising:

[0020] The first obtaining module is used to obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0021] The second obtaining module is used to obtain at least one image feature of the candidate image; wherein the image feature is a feature that can characterize the imaging quality of the content to be identified;

[0022] The result determination module is used to determine the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the at least one image feature.

[0023] Fifthly, embodiments of this application provide an image filtering device, comprising:

[0024] The recognition result acquisition module is used to obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0025] An image content recognition module is used to perform image recognition processing on a reference image of the candidate image to obtain the reference content of the candidate image; wherein the content to be recognized in the reference image is the same as the content to be recognized in the candidate image.

[0026] The indicator analysis module is used to determine the target auxiliary evaluation indicators of the candidate image based on the reference content of the candidate image.

[0027] The filtering result determination module is used to determine the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the target auxiliary evaluation index.

[0028] Sixthly, embodiments of this application provide an identification device applied to a client, the device comprising:

[0029] The receiving module is used to receive configuration requests for samples;

[0030] The display module is used to respond to the configuration request and display a first recognition result and a first evaluation index for the sample; the first evaluation index indicates the imaging quality of the corresponding first recognition result; the first recognition result is the recognition result obtained by image recognition processing of multiple imaging results of the sample, and the corresponding first evaluation index satisfies the first screening condition; the multiple imaging results are obtained by imaging the sample based on set multi-group imaging conditions.

[0031] The setting module is used to respond to user interaction and set the imaging conditions corresponding to the first recognition result to the first imaging conditions for the sample to complete the configuration.

[0032] In a seventh aspect, embodiments of this application provide an electronic device, including:

[0033] Memory, used to store computer programs;

[0034] The processor, when executing a program stored in memory, implements any of the described image filtering or recognition methods.

[0035] Eighthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements any of the described image filtering or recognition methods.

[0036] Ninthly, embodiments of this application also provide a computer program product containing instructions that, when run on a computer, cause the computer to perform any of the image filtering or recognition methods described above.

[0037] Beneficial effects of the embodiments in this application:

[0038] In an image filtering method provided in this application embodiment, for at least one candidate image, a recognition result obtained through image recognition processing is acquired, and at least one image feature of the candidate image is obtained, wherein the image feature is a feature capable of characterizing the imaging quality of the content to be recognized; thereby, using the recognition result of the candidate image and the at least one image feature as filtering criteria, an image filtering result is determined from the at least one candidate image. It can be seen that this solution, based on the recognition result of image recognition processing, combines features capable of characterizing the imaging quality of the content to be recognized to perform image filtering, making the filtering criteria more diverse. Compared with existing technologies, this reduces the influence of the recognition effect of image recognition processing on image filtering, thereby improving the accuracy of image filtering.

[0039] In another image filtering method provided in this application embodiment, for at least one candidate image, a recognition result obtained by image recognition processing is obtained. Furthermore, image recognition processing is performed on a reference image of the candidate image to obtain reference content of the candidate image. Based on the reference content of the candidate image, a target auxiliary evaluation index for the candidate image is determined. Finally, based on the recognition result and the target auxiliary evaluation index of the candidate image, an image filtering result is determined from the at least one candidate image. It can be seen that this scheme performs image filtering based on the recognition result and the target auxiliary evaluation index determined based on the reference content in the reference image. This allows for diverse filtering criteria without relying on the ground truth of the candidate image. Compared to existing technologies, it reduces the impact of the recognition effect of image recognition processing on image filtering, thereby improving the accuracy of image filtering.

[0040] The identification method provided in this application embodiment allows the client to receive a configuration request for a sample, and in response to the configuration request, display a first identification result and a first evaluation index for the sample; and, in response to user interaction, set the imaging conditions corresponding to the first identification result as the first imaging conditions for the sample to complete the configuration. It is evident that in this solution, through simple interaction between the user and the client, a configuration process for the first imaging conditions of the sample is added, thereby enabling the online identification process for objects of the same type as the sample to reduce the number of imaging conditions required while ensuring identification accuracy, providing a data foundation for improving identification efficiency; furthermore, through simple interaction between the user and the client, the configuration of imaging conditions for the sample can be completed, thereby lowering the operational threshold, simplifying the configuration process, and improving configuration efficiency and user experience.

[0041] Of course, implementing any product or method of this application does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0043] Figure 1 This is a schematic diagram illustrating the selection of recommended images from a candidate imaging queue.

[0044] Figure 2 A flowchart illustrating an image filtering method provided in an embodiment of this application;

[0045] Figure 3 This is another flowchart of an image filtering method provided in an embodiment of this application;

[0046] Figure 4 A flowchart illustrating another image filtering method provided in this application embodiment;

[0047] Figure 5 A flowchart illustrating an imaging recommendation method provided in an embodiment of this application;

[0048] Figure 6 Another flowchart of an imaging recommendation method provided in an embodiment of this application;

[0049] Figure 7 A flowchart illustrating an identification method provided in an embodiment of this application;

[0050] Figure 8 This is another flowchart of an identification method provided in an embodiment of this application;

[0051] Figure 9 A schematic diagram of the client interface provided in an embodiment of this application;

[0052] Figure 10 This is another schematic diagram of the client interface provided in the embodiments of this application;

[0053] Figure 11 A schematic diagram illustrating the principle of an identification method provided in an embodiment of this application;

[0054] Figure 12 This is a schematic diagram of the structure of an image filtering device provided in an embodiment of this application;

[0055] Figure 13 This is a schematic diagram of another image filtering device provided in an embodiment of this application;

[0056] Figure 14 This is a schematic diagram of the structure of an identification device provided in an embodiment of this application;

[0057] Figure 15 This is a schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0058] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of this application.

[0059] To better understand the image filtering method provided in the embodiments of this application, the following section will first introduce the relevant content of imaging recommendation (the image filtering process is one of the processing stages).

[0060] The image polling function based on the recognition effect of image recognition processing (also known as the image polling service) can include two stages: image acquisition process and image recommendation. The so-called image recommendation refers to the recommendation of imaging configuration information, which can include one or more of the following information that can affect the imaging effect: shooting parameters (such as focal length, ISO, aperture, shooting mode, etc.), imaging algorithm, lighting conditions (such as light source type, light intensity, color temperature, etc.).

[0061] For example, the implementation process of this imaging polling function may include: image acquisition, image recognition processing (such as text recognition), quality evaluation, and result recommendation; specifically, the implementation process may include: setting different imaging configuration information (such as lighting conditions, camera parameters, and / or imaging algorithms, etc.) manually or automatically to capture a series of images of the same target object (such as the same target object containing character content); evaluating the series of images based on the recognition results (such as text recognition results, also known as character recognition results), and using the image with the highest score as the image selection result; determining the imaging recommendation result that at least contains imaging configuration information based on the imaging configuration information when the image selection result was captured; such as Figure 1 As shown, a recommended image is selected from the candidate image queue, which can be a series of images, and the recommended image is the image with the highest score.

[0062] Furthermore, imaging recommendation in real-world scenarios can be divided into two usage stages: offline imaging recommendation (also known as offline configuration recommendation) and online imaging recommendation (also known as online recipe recommendation); that is, for imaging recommendation, there are two types: offline and online, and these two imaging recommendation methods are implemented based on the imaging polling function. Specifically:

[0063] Offline imaging recommendation refers to: for a specific type of sample (e.g., a target object with known type and containing character content), selecting the best imaging configuration from all imaging configurations (i.e., imaging configuration information) (e.g., imaging configuration information that ensures the character image quality meets quality requirements when capturing an image, such as clear character image), and binding the selected imaging configuration with that specific type as an imaging "recipe" to obtain the imaging recommendation result; of course, the above imaging configuration information (i.e., imaging conditions) can be called a recipe, and through offline imaging recommendation, a recipe that meets the imaging quality requirements is set for each sample (i.e., selecting the best recipe).

[0064] Online imaging recommendation refers to recommending the best imaging "recipe" (of course, in some cases, the image using the best imaging "recipe") from a series of imaging "recipes" obtained from the preceding offline configuration recommendation for samples of uncertain types (such as a target object of unknown type containing character content) during the actual production process, for subsequent processing.

[0065] It is evident that the difference between these two usage phases lies in the following: for offline imaging recommendation, more prior information is known (including sample type and string truth values), the configuration options are wider, and manual adjustment is generally allowed; while for online imaging recommendation, the best "recipe" is selected from a limited set of imaging "recipes," and there is no prior information, relying entirely on the evaluation of recognition performance.

[0066] Furthermore, the imaging polling and imaging recommendation process has the following characteristics:

[0067] In the image screening stage, the image recognition process relies solely on the recognition effect of the image recognition process, which leads to the imaging recommendation results depending only on the recognition effect of the image recognition process (such as the text recognition effect).

[0068] It is necessary to manually set the region information of the area to be recognized for image recognition processing (such as setting the text recognition region and specifying the size of a single character).

[0069] For text recognition, it can only recognize font types supported by a limited font configuration.

[0070] The above three characteristics limit the applicability and ease of use of imaging recommendation methods, restricting their application to limited industry scenarios and requiring highly skilled personnel for configuration and operation. Furthermore, since imaging recommendation results rely solely on the recognition performance of image recognition processing, there is a risk that the recommended optimal imaging configuration may not be found if there is a discrepancy between the specific implementation method of image recognition processing and human evaluation. Additionally, when encountering unrecognizable content, neither valid results can be identified, nor can usable imaging recommendations be made.

[0071] It is evident that reducing the impact of image recognition processing on image selection, thereby improving the accuracy of image selection, is crucial for enhancing the accuracy of image recommendation.

[0072] Based on the above description, this application provides an image filtering method to reduce the impact of image recognition processing on image filtering, thereby improving the accuracy of image filtering.

[0073] The following describes an image filtering method provided by an embodiment of this application.

[0074] The image filtering method provided in this application can be applied to electronic devices. In specific applications, the electronic device can be a terminal device or a server. For example, if the electronic device is a terminal device, it can be a data acquisition device with image acquisition function, or a processing device that communicates with the data acquisition device, etc.

[0075] In addition, this image filtering method can be applied to application scenarios involving imaging recommendation. Specifically, this image filtering method can be applied to the image filtering stage in imaging recommendation. Furthermore, application scenarios involving imaging recommendation can include, but are not limited to, industrial workpiece defect detection, industrial character and label recognition, production line packaging quality inspection, logistics warehousing and sorting, precision electronic manufacturing, heavy metal inspection, factory security monitoring and other industrial vision scenarios.

[0076] Of course, this image filtering method can also be applied to processing scenarios with image filtering requirements (e.g., for image filtering of the same target object, where the same target object itself can be the content to be identified, or the character content contained in the same target object can be the content to be identified). For example, this image filtering method can be applied to image acquisition scenarios of target objects containing character content. In this image acquisition scenario, at least one image can be captured for the target object to serve as candidate images. Based on this image filtering method, images with character imaging quality that meet the quality requirements (e.g., clear character imaging) are selected from these at least one candidate image and used as the image acquisition result for subsequent image processing.

[0077] It should be emphasized that the embodiments of this application do not limit the specific scenarios in which the image filtering method is applied. Any scenario with image filtering requirements can be used as the application scenario for the image filtering method provided in the embodiments of this application.

[0078] The image filtering method provided in this application embodiment may include the following steps:

[0079] Obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0080] Obtain at least one image feature of the candidate image; wherein the at least one image feature is a feature capable of characterizing the image quality of a character;

[0081] Based on the recognition results of the candidate images and the at least one image feature, an image filtering result is determined from the at least one candidate image.

[0082] In the solution provided in this application embodiment, for at least one candidate image, a recognition result obtained through image recognition processing is obtained, and at least one image feature of the candidate image is obtained, wherein the image feature is a feature that can characterize the imaging quality of the content to be recognized; thereby, using the recognition result of the candidate image and the at least one image feature as the screening criteria, an image screening result is determined from the at least one candidate image. It can be seen that this solution, based on the recognition result of image recognition processing, combines features that can characterize the imaging quality of the content to be recognized to perform image screening, making the screening criteria more diverse. Compared with existing technologies, this can reduce the influence of the recognition effect of image recognition processing on image screening, thereby improving the accuracy of image screening.

[0083] The image filtering method provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0084] like Figure 2 As shown in the embodiments of this application, an image filtering method may include:

[0085] S201, Obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0086] S202, obtain at least one image feature of the candidate image; wherein, the image feature is a feature that can characterize the imaging quality of the content to be identified;

[0087] S203, based on the recognition result of the candidate image and the at least one image feature, determine the image filtering result from the at least one candidate image.

[0088] In this embodiment, given that at least one candidate image has been determined, there is no strict execution order between S201 and S202. For example, S201 can be executed before S202, or after S202. Of course, S201 and S202 can also be executed in parallel. In specific applications, the number of candidate images can be at least two, but is not limited to this. The same processing procedure is performed for each candidate image.

[0089] The at least one candidate image can be any image captured for the same target object based on different imaging configuration information. For example, if image filtering is used to select images with higher character imaging quality, then the content to be identified in the target object can be the character content contained in the target object, and the corresponding image recognition processing can be text recognition; or, if image filtering is used to select images with higher imaging quality in the target object itself, then the content to be identified can also be the target object itself, and the image recognition processing can be object classification (also known as target classification), object segmentation, object anomaly detection, or other object recognition processing. The candidate image can be an image captured based on imaging configuration information. The method of obtaining the at least one candidate image can vary depending on the application scenario, such as different shooting devices, different imaging configuration information used, different target objects captured, etc. This application does not specifically limit the method of obtaining the at least one candidate image.

[0090] Furthermore, when the content to be identified is character content, the target object containing character content can exist in various forms. For example, the target object can be an industrial workpiece, paper carrier, electronic display interface, signage, or any other carrier or object that carries text, numbers, symbols, and / or codes. Additionally, when the content to be identified is the target object itself, the target object can be any object that requires identification, such as a workpiece, material product, or finished product. Moreover, in different application scenarios, the type of the target object can be known or unknown.

[0091] After obtaining at least one candidate image, since the purpose of image screening is to select images whose imaging quality meets the requirements, image recognition processing can be performed on the candidate image to obtain the recognition result obtained from the image recognition processing of at least one candidate image. This processing result can then be used as a factor in image screening, i.e., as a screening criterion. For example, if the content to be recognized is character content, the recognition result can include the character content and its confidence level. The confidence level can be the confidence level of each individual character or the overall confidence level of the character content; both are reasonable. For example, if the content to be recognized is the target object itself, the recognition result can be the object recognition result for the target object and its confidence level. Furthermore, this application embodiment does not limit the specific recognition method for image recognition processing; any recognition method capable of image recognition processing can be applied to this application embodiment.

[0092] Furthermore, to reduce the impact of image recognition processing on the recognition effect, in this embodiment, for at least one candidate image, in addition to image recognition processing, at least one image feature is extracted. This at least one image feature is used as a factor to consider in image screening, that is, as a screening criterion other than the recognition result of image recognition processing. It is understood that when screening images based on imaging effects (such as character imaging effects) through manual recognition (human eye recognition), the color dimension features of candidate images are more favored. Therefore, by using this at least one image feature as an additional screening criterion, the effect recognized by both automatic recognition and human eye recognition can be taken into account.

[0093] It should be noted that the candidate image may include the recognition region of the content to be recognized (such as the character region or the object region of the target object), as well as other object regions outside the recognition region. Therefore, in order to ensure that the at least one image feature can accurately reflect the imaging quality of the content to be recognized (such as the clarity of characters or the clarity of the target object), the at least one image feature of the candidate image may specifically be: at least one image feature of the recognition region in the candidate image. The recognition region may be the region located by the image recognition processing or the region specified manually.

[0094] The at least one image feature extracted from the candidate image is a feature that can characterize the imaging quality of the content to be identified (such as character imaging quality). For example, in one implementation, the at least one image feature of the candidate image may include, but is not limited to, one or more of the following: average gray level, gray level variance, contrast, average gradient, and gray level moment. The specific image feature to be extracted can be manually set according to the actual situation or determined by a predetermined determination method. Furthermore, the specific extraction method for each image feature in this application embodiment may include: feature extraction through a pre-trained neural network model, or extraction through other set feature extraction methods. This application does not limit the specific extraction method, and any method that can extract the at least one image feature can be applied to the embodiments of this application. For example, in another implementation, the at least one image feature includes at least one feature extracted from the candidate image based on a preset feature extraction model; the preset feature extraction model is a model trained based on image samples and corresponding label information; wherein, the label information is used to characterize the imaging quality of the content to be identified in the image sample (for example, imaging quality can be the clarity of characters, and for clarity, it can be divided into at least two levels, such as clear and blurry, or high-definition, medium-definition, and low-definition); by training the model based on image samples and corresponding label information, the feature extraction model can learn image features that can characterize different imaging qualities. It should be emphasized that the image features mentioned in the above two implementations can be combined to obtain at least one image feature of the candidate image.

[0095] After obtaining the recognition result and the at least one image feature of the candidate image, the recognition result and the at least one image feature are used as screening criteria to evaluate the at least one candidate image, so as to determine the image screening result from the at least one candidate image based on the evaluation.

[0096] There are multiple ways to determine the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the at least one image feature.

[0097] Optionally, in one implementation, determining the image filtering result from the at least one candidate image based on the recognition result of the candidate image and at least one image feature may include:

[0098] Based on the recognition results of the candidate images and at least one image feature, the imaging quality of the candidate images is scored to obtain the quality score value of each of the at least one candidate image; and, according to the determined quality score values, the candidate image with the highest quality score value is selected from the at least one candidate image to obtain the image filtering result.

[0099] For example, based on the recognition results of the candidate images and at least one image feature, the imaging quality of the candidate images is scored to obtain a quality score value for each of the at least one candidate image, which may include:

[0100] Based on at least one evaluation metric of the candidate image, a score value for at least one scoring dimension of the candidate image is determined, and based on the score value for at least one scoring dimension of the candidate image, a quality score value for the imaging quality of the candidate image is analyzed to obtain a quality score value for each of the at least one candidate image; wherein, the at least one evaluation metric of the candidate image includes: the confidence level in the recognition result of the candidate image and at least one metric among at least one image feature of the candidate image; wherein, confidence level and image feature can both be used as metric.

[0101] For example, a scoring dimension can uniquely correspond to an evaluation index, and the confidence level in the recognition result of the candidate image can be used as an evaluation index, with each image feature serving as an evaluation index. Each evaluation index can have its own scoring strategy. Thus, for each evaluation index of the candidate image, the score value of the corresponding scoring dimension is determined according to the scoring strategy corresponding to that evaluation index.

[0102] It is understandable that, for each evaluation index belonging to image features, if the larger the value of the evaluation index, the better the image quality, then the scoring strategy set for the evaluation index may include a strategy where the index value and the score value are positively correlated; if the smaller the value of the evaluation index, the worse the image quality, then the scoring strategy set for the evaluation index may include a strategy where the index value and the score value are negatively correlated. For evaluation indices belonging to confidence, the scoring strategy set for the evaluation index may include a strategy where the confidence level and the score value are positively correlated. This application embodiment does not specifically limit the specific content of the scoring strategy set for each evaluation index. For example, in other implementations, for any evaluation index to which an index type belongs, a correspondence between candidate index content and score values ​​for that index type can be established. Then, for a candidate image, based on this correspondence, candidate index content that matches the index content of that evaluation index of the candidate image is selected, and the score value corresponding to the selected candidate index content is used as the score value corresponding to that evaluation index of the candidate image.

[0103] In addition, the method of analyzing the imaging quality score of the candidate image based on the score value of at least one scoring dimension of the candidate image may include, but is not limited to: directly weighting each score value, performing mathematical operations (squaring or taking the square root) on each score value and then weighting it, etc.

[0104] The specific implementation of scoring the imaging quality of the candidate images based on the recognition results of the candidate images and at least one image feature to obtain the quality score value of each of the at least one candidate image is merely an example and should not constitute a limitation on the embodiments of this application.

[0105] Furthermore, the aforementioned implementation method of determining the image filtering result from the at least one candidate image based on the recognition result of the candidate image and at least one image feature is merely an example and should not constitute a limitation on the embodiments of this application. For example, in other implementation methods, confidence level is used as a primary filtering indicator, and the at least one image feature is used as a secondary filtering indicator. Specifically, from at least one candidate image, candidate images with a confidence level (e.g., the confidence level of each character or the overall confidence level of the character content) higher than a predetermined confidence level threshold are selected. Then, for each of the selected candidate images, each image feature is used as an evaluation indicator to score the imaging quality of the candidate image, thereby obtaining the quality score value of the candidate image; thus, from the obtained candidate images, the candidate image with the highest quality score value is selected to obtain the image filtering result.

[0106] In the solution provided in this application embodiment, for at least one candidate image, a recognition result obtained through image recognition processing is acquired, and at least one image feature of the candidate image is obtained, wherein the image feature is a feature capable of characterizing the imaging quality of the content to be recognized; thereby, using the recognition result of the candidate image and the at least one image feature as the screening criteria, an image screening result is determined from the at least one candidate image. It can be seen that this solution, based on the recognition result and combined with features capable of characterizing imaging quality, performs image screening, making the screening criteria more diverse. Compared with existing technologies, this reduces the influence of the recognition effect of image recognition processing on image screening, thereby improving the accuracy of image screening.

[0107] In one possible embodiment, step S202 described above may include:

[0108] Step A: According to the target feature extraction strategy, extract image features from the recognition region of the candidate image to obtain at least one image feature of the candidate image; wherein, the target feature extraction strategy is used to characterize the type of image feature to be extracted and / or the feature extraction method used.

[0109] In this embodiment, for the candidate image, a target feature extraction strategy can be determined first, that is, the types of image features to be extracted and / or the feature extraction methods to be used can be clearly defined. Then, according to the determined target feature extraction strategy, image features are extracted from the recognition region of the candidate image in a targeted manner to obtain at least one image feature of the candidate image. To ensure the effective evaluation of the candidate image, the at least one candidate image uses the same target feature extraction strategy.

[0110] It is understandable that when the target feature extraction strategy only represents the type of image features to be extracted, the type of image features to be extracted is fixed, while the specific feature extraction method can be any method; when the target feature extraction strategy only represents the feature extraction method used for the image features to be extracted, the type of image features to be extracted can be a default type; and when the target feature extraction strategy represents both the type of image features to be extracted and the feature extraction method used, the type of image features to be extracted and the feature extraction method used are both fixed.

[0111] Furthermore, there can be multiple ways to determine the target feature extraction strategy. For example, in one implementation, the current feature extraction strategy is read from a pre-defined storage area for feature extraction strategies to determine the target feature extraction strategy; the feature extraction strategies stored in the storage area can be updated when update conditions are met, for example, the update conditions can be a setting instruction for receiving a feature extraction strategy.

[0112] In another implementation, a manually configured feature extraction strategy can be obtained through the configuration interface corresponding to the at least one candidate image, thereby obtaining the target feature extraction strategy.

[0113] In this embodiment, for the candidate image, image features can be extracted from the recognition region of the candidate image according to the target feature extraction strategy to obtain at least one image feature of the candidate image; thus, the extraction of image features of the candidate image is more flexible and targeted, thereby providing a basis for improving the accuracy of image screening.

[0114] In one possible embodiment, the method for determining the target feature extraction strategy may include:

[0115] Step B1: If target additional information is set for the at least one candidate image, a target feature extraction strategy is generated based on the target additional information; wherein, the target additional information represents: morphological features and / or content information about the content to be identified;

[0116] Step B2: If no additional target information is set for the target candidate image, the default feature extraction strategy is determined as the target feature extraction strategy.

[0117] To determine a target feature extraction strategy suitable for the at least one candidate image, the presence or absence of target supplementary information is used as the analysis criterion. Different determination methods are employed to obtain the target feature extraction strategy for the at least one candidate image. Specifically, when target supplementary information exists, it indicates that guidance information regarding the target feature extraction strategy has been provided for the at least one candidate image. Therefore, based on the target supplementary information, a feature extraction strategy adapted to the at least one candidate image is generated to obtain the target feature extraction strategy. Conversely, when target supplementary information does not exist, it indicates that no guidance information regarding the target feature extraction strategy has been provided for the at least one candidate image. Therefore, the default feature extraction strategy is determined as the target feature extraction strategy to ensure that a usable feature extraction strategy can be obtained. The default feature extraction strategy is a pre-defined, universally applicable feature extraction strategy.

[0118] For example, the target supplementary information may include at least one of the morphological features and content information of the content to be identified in the at least one candidate image. Taking character content as an example, the morphological features of the content to be identified are character morphological features, which may include, but are not limited to, one or more of the following: font type, character shape, and size of the recognition area. Taking the target object itself as an example, the morphological features of the content to be identified can be object morphological features, which may include, but are not limited to, one or more of the following: overall outline, local outline, surface texture, geometric dimensions, relative position of components, edge curvature, surface concavity and convexity, color distribution, and defect deformation features. It is understood that the character shape can be a shape defined structurally (i.e., stroke structure dimension), such as a dot matrix shape or a continuous shape, or a shape defined technologically (i.e., molding process dimension), such as a stamped shape or a printed shape (e.g., color printing shape).

[0119] It should be noted that the shape features of the content to be identified can change the foreground pixel ratio, texture complexity, and edge jaggedness of the recognition area. Therefore, while ensuring the effectiveness of the extracted image features, different shape features can be adapted to different image features and / or feature extraction methods. For example, font type and character shape will change the foreground pixel ratio, stroke texture complexity, and edge jaggedness of the recognition area, respectively. Therefore, while ensuring the effectiveness of the extracted image features, different font types and different character shapes can be adapted to different image features and / or feature extraction methods.

[0120] Furthermore, different content information (such as specific character content) also has imaging differences. For example, the imaging texture, grayscale, and edge distribution may be different. Also, different character content may have different noise resistance under the same imaging environment. Therefore, content information can also be considered as a factor when determining the feature extraction strategy.

[0121] Based on this, taking character shape features as an example, a first type of mapping relationship can be pre-established between the feature content of various character shape features and the types of image features to be extracted, and / or a second type of mapping relationship between the feature content of various character shape features and the feature extraction methods used by the image features to be extracted; thus, when the target additional information contains character shape features, based on the first type of mapping relationship and / or the second type of mapping relationship, the types of image features to be extracted and / or the feature extraction methods used are matched, thereby generating a target feature extraction strategy.

[0122] For example, the first type of mapping relationship may include: a well-structured sans-serif font corresponds to average gray level, contrast and average gradient; a textured Song or Kai font corresponds to gray level variance, gray level moment and average gradient; a smooth vector font (i.e. a continuous font) corresponds to average gray level and contrast; and a dot matrix font (i.e. a dot matrix font) corresponds to features based on edges or gray level gradients.

[0123] Furthermore, the size of the recognition region changes the proportion of foreground pixels and edge pixels of the text. Therefore, the size of the recognition region can be used to constrain the types of image features to be extracted and / or the feature extraction method used, thereby ensuring the effectiveness of the extracted image features.

[0124] Based on this, a third type of mapping relationship can be pre-established between the size information of the recognition region and the types of image features to be extracted, and / or a fourth type of mapping relationship between the size information of the recognition region and the feature extraction method used. Thus, when the target supplementary information includes the size information of the recognition region, the types of image features to be extracted and / or the feature extraction method used can be matched based on the third type and / or fourth type of mapping relationship, thereby generating a target feature extraction strategy. For example, recognition regions larger than a set size threshold have sufficient pixel information and a stable overall grayscale distribution, allowing the extraction of global statistical features such as average grayscale, contrast, and grayscale moments. Recognition regions smaller than the set threshold have fewer effective pixels and a higher proportion of edges, making global grayscale statistical features susceptible to noise interference and ineffective. Therefore, features such as average gradient and grayscale variance, which focus on local edges and discrete grayscale changes, can be extracted. Based on this, the third type of mapping relationship can include: size information larger than a predetermined size threshold corresponds to average grayscale, contrast, and grayscale moments; size information smaller than the predetermined size threshold corresponds to average gradient and grayscale variance.

[0125] In this embodiment, the target additional information serves as guidance information for the type of image features to be extracted and / or the feature extraction algorithm to be used. Furthermore, based on the analysis results of whether the target additional information exists, a target feature extraction strategy is obtained through different determination methods. This ensures that the target feature extraction strategy is obtained and is more compatible with the candidate image, thus providing a foundation for the effective extraction of image features.

[0126] In one possible implementation, such as Figure 3 As shown, an image filtering method may include the following steps:

[0127] S301, Obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0128] S302, obtain at least one image feature of the candidate image; wherein, the image feature is a feature that can characterize the imaging quality of the content to be identified;

[0129] S303, Obtain the reference content of the candidate image;

[0130] The reference content is used to evaluate the accuracy of the recognition results of the candidate images.

[0131] S304, Based on the reference content of the candidate image, determine the target auxiliary evaluation index of the candidate image;

[0132] The target auxiliary evaluation index includes a first auxiliary evaluation index and / or a second auxiliary evaluation index; the first auxiliary evaluation index characterizes the degree of consistency between the result content represented by the recognition result of the candidate image and the reference content; the second auxiliary evaluation index characterizes the confidence level of the reference content of the candidate image.

[0133] S305, based on the recognition result of the candidate image, the at least one image feature, and the target auxiliary evaluation index, determine the image screening result from the at least one candidate image.

[0134] In this embodiment, S301-S302 are the same as S201-S202 in the above embodiment, and will not be described again here; and, provided that at least one candidate image is determined, there is no strict execution order between S303 and S301, S302. For example, S301, S302 and S303 can be executed in parallel.

[0135] To further improve the accuracy of image filtering, additional filtering criteria (considerations) can be added to the recognition results and at least one image feature to enhance the diversity of filtering criteria. Specifically, considering that the accuracy of the content in the recognition results of each candidate image (such as the accuracy of character content or object category in the recognition results) is a major consideration dimension for image filtering, reference content can be obtained for each candidate image. This reference content, which can evaluate the accuracy of the recognition results of the candidate image, further reflects the accuracy of the content in the recognition results of the candidate image. Therefore, based on the recognition results, at least one image feature, and reference content of each candidate image, the image filtering results can be determined from the candidate images.

[0136] The reference content of the candidate image can be the truth value corresponding to the at least one candidate image (such as the text truth value of the target object), or other content that can be used to judge the degree of accuracy. Since the at least one candidate image may or may not have a truth value in different application scenarios, a predetermined determination method can be used to determine the reference content of the candidate image, so as to flexibly respond to the analysis results of whether a truth value exists. For the sake of clarity and layout, the specific determination method will be set in subsequent embodiments.

[0137] Furthermore, for the candidate image, a target auxiliary evaluation index (including only the first auxiliary evaluation index, or including both the first and second auxiliary evaluation indices) can be determined based on the reference content of the candidate image and the character content represented by the recognition result of the candidate image. After obtaining the target auxiliary evaluation index of the candidate image, the candidate image can be evaluated based on the recognition result of the candidate image, the at least one image feature, and the target auxiliary evaluation index to determine the image screening result from the at least one candidate image, thereby performing image screening through multi-dimensional index information.

[0138] The similarity (or consistency) between the recognition result of the candidate image and the reference content reflects the accuracy of the recognition result. Therefore, similarity can be used as an additional screening criterion for image selection. The specific calculation method for the similarity between the recognition result of the candidate image and the reference content can include, but is not limited to, calculations based on at least one of the following indicators: edit distance, number of common characters, and total length of common substrings. It is understood that different indicators have different tolerances for differences between strings, and the indicator can be selected according to application requirements during specific applications.

[0139] Furthermore, the confidence level of the reference content of the candidate image reflects the reliability of the aforementioned similarity for quality analysis. Therefore, the confidence level of the reference content of the candidate image can be used as an additional analytical basis for image screening. It is understood that if the reference content of the candidate image is the true value corresponding to the candidate image, then the confidence level of the reference content of the candidate image can be 1; if the reference content of the candidate image is not the true value corresponding to the candidate image, i.e., it is other content that can be used to assess accuracy, a confidence level less than 1 can be set for the reference content of the candidate image. Alternatively, if the reference content of the candidate image is content obtained from an image through image recognition processing, the confidence level obtained from the image recognition processing can be used as the confidence level of the reference content of the candidate image.

[0140] There are multiple ways to determine the image selection result from the at least one candidate image based on the recognition result of the candidate image, the at least one image feature, and the target auxiliary evaluation index.

[0141] For example, in one implementation, based on the recognition result of the candidate image, the at least one image feature, and the target-assisted evaluation index, the imaging quality of the candidate image with respect to the content to be identified is scored to obtain a quality score value for the candidate image; and, according to the determined quality score values, the candidate image with the highest quality score value is selected from the at least one candidate image to obtain the image filtering result. The specific implementation process of this method will be described exemplarily in the embodiments below.

[0142] For example, in other implementations, confidence level is used as the primary screening metric, and the at least one image feature and the target auxiliary evaluation metric are used as secondary screening metrics. Specifically, from the at least one candidate image, candidate images with a confidence level (e.g., the confidence level of each character or the overall confidence level of the character content) higher than a predetermined confidence level threshold are selected. Then, for each of the selected candidate images, each image feature and each auxiliary evaluation metric in the target auxiliary evaluation metric are used as an evaluation metric to score the imaging quality of the candidate image, thereby obtaining the quality score value of the candidate image. Thus, from the selected candidate images, the candidate image with the highest quality score value is selected to obtain the image screening result.

[0143] The above-described identification results, at least one image feature, and target auxiliary evaluation index based on the candidate images, used to determine the image filtering results from at least one candidate image, are merely examples and should not constitute a limitation on the embodiments of the application.

[0144] In this embodiment, the target auxiliary evaluation index of the candidate image is determined by the reference content of the candidate image; and based on the recognition result of the candidate image, the at least one image feature, and the target auxiliary evaluation index, the image screening result is determined from the at least one candidate image; thus, based on the recognition result of image recognition processing, combined with features that can characterize imaging quality and the target auxiliary evaluation index, image screening is performed, making the screening criteria more diverse. Compared with the prior art, this can reduce the influence of the recognition effect of image recognition processing on image screening, thereby improving the accuracy of image screening.

[0145] In one possible embodiment, S303 described above may include:

[0146] Step C1: If a truth value exists for the candidate image, the truth value is determined as the reference content of the candidate image;

[0147] Step C2: If the true value does not exist, obtain the reference image of the candidate image, and perform image recognition processing on the reference image of the candidate image to obtain the reference content of the candidate image;

[0148] The content to be identified in the reference image of the candidate image is the same as the content to be identified in the candidate image.

[0149] In the process of obtaining reference content, ground truth (e.g., textual ground truth) can be preferentially selected as reference content. Based on this processing idea, it is possible to first analyze whether there is a ground truth corresponding to the candidate image (e.g., whether there is a textual ground truth in the at least one candidate image), and perform different processing procedures according to different analysis results. Specifically, if there is a ground truth, the ground truth is determined as the reference content of the candidate image to ensure the accuracy of the reference content; if there is no ground truth, the reference image of the candidate image is obtained, and image recognition processing is performed on the reference image corresponding to the candidate image to obtain the reference content of the candidate image. Thus, in the absence of a ground truth, content other than the ground truth is introduced as reference content to evaluate the recognition result of the candidate image in the absence of a ground truth.

[0150] There are multiple ways to analyze whether a truth value exists for the candidate image. For example, in one implementation, the existence of a truth value for the candidate image can be analyzed based on the indication information set through the configuration interface to indicate whether a truth value exists.

[0151] For example, in other implementations, analyzing whether a truth value exists corresponding to the candidate image may include:

[0152] If the at least one candidate image is an image required in the first type of image task, then the existence of a ground truth value corresponding to the candidate image is analyzed.

[0153] If the at least one candidate image is the selected image required in the second type of image task, then it is analyzed that there is no true value corresponding to the candidate image;

[0154] The first type of image task is an image task with a set truth value, and the second type of image task is an image task without a set truth value.

[0155] Considering that some types of image tasks are tasks with a set truth value, while others are tasks without a set truth value, it is possible to analyze which image task the at least one candidate image is required for, thereby determining whether a truth value corresponding to the candidate image exists.

[0156] Specifically: if the at least one candidate image is an image required in the first type of image task, that is, an image required in an image task with a set truth value, then it is analyzed that a truth value corresponding to the candidate image exists; if the at least one candidate image is an image required in the second type of image task, that is, an image required in an image task without a set truth value, then it is analyzed that a truth value corresponding to the candidate image does not exist.

[0157] It is understood that the method of analyzing whether the at least one candidate image is an image required for the first type of image task or an image required for the second type of image task is not limited in the embodiments of this application.

[0158] To ensure clarity of the solution and layout, we will analyze whether there are other ways to implement the truth value corresponding to the at least one candidate image, which will be introduced later in conjunction with other embodiments.

[0159] Furthermore, there can be multiple ways to obtain the reference image for the candidate image. For example, in one implementation, one image can be selected from the at least one candidate image according to a predetermined selection strategy, and this predetermined selection strategy can include, but is not limited to, a random selection strategy, a strategy of selecting the candidate image with the highest similarity in imaging configuration information, etc. For clarity and layout, other implementations of obtaining the reference image for the candidate image will be described later in conjunction with other embodiments.

[0160] Furthermore, this application does not limit the specific method of performing image recognition processing on the reference image of the candidate image; any method capable of performing image recognition processing can be applied to the embodiments of this application.

[0161] In this embodiment, the existence of a ground truth value corresponding to at least one candidate image is analyzed, and different processing procedures are performed based on different analysis results to obtain the reference content of the candidate image. Therefore, the reference content of the candidate image can be flexibly obtained based on the analysis results of whether a ground truth value corresponding to at least one candidate image exists, making the reference content acquisition process versatile.

[0162] In one possible embodiment, the reference image for obtaining the candidate image may include:

[0163] The candidate image is subjected to image transformation processing, and a reference image is determined based on the image obtained after image transformation processing.

[0164] Image transformation processing may include, but is not limited to, one or more of the following: grayscale transformation processing, pixel transformation processing, and several transformation processing.

[0165] It is understood that grayscale transformation processing is used to map and adjust the grayscale value range of pixels; for example, grayscale transformation processing may include, but is not limited to: linear grayscale transformation (such as linear stretching / compression), nonlinear grayscale transformation (such as logarithmic transformation, exponential transformation, power transformation), grayscale inversion (such as negative transformation), piecewise linear grayscale transformation (such as piecewise stretching, piecewise contrast adjustment), histogram equalization, grayscale thresholding (such as binarization, multi-level thresholding), gamma correction (Gamma transformation), brightness adjustment, contrast adjustment, etc.

[0166] Pixel transformation processing is used to perform spatial filtering, interpolation, image enhancement, or noise reduction on pixels and their neighborhoods in candidate images. For example, pixel transformation processing may include, but is not limited to: image smoothing filtering (such as mean filtering, Gaussian filtering, median filtering, or bilateral filtering), image sharpening (such as gradient sharpening, Laplacian sharpening, or edge enhancement), pixel interpolation operations (such as nearest neighbor, bilinear, bicubic interpolation), pixel filling, pixel cropping, pixel addition, deletion, and edge padding, noise removal, etc.

[0167] Geometric transformation processing is used to transform the spatial coordinates of candidate images; for example, geometric transformation processing may include, but is not limited to: changing aspect ratio, distorting, tilting, translating, rotating, scaling, flipping, etc.

[0168] For example, after performing image transformation processing on the candidate image, if only one image transformation processing result is obtained, the obtained image is directly determined as the reference image of the candidate image; if at least two image transformation processing results are obtained, image fusion processing can be performed on the at least two images, and the fused image is determined as the reference image of the candidate image.

[0169] In this embodiment, by performing image transformation processing on the candidate image, a reference image can be determined based on the image obtained after the image transformation processing. This ensures a high correlation between the reference image and the candidate image, providing a foundation for obtaining more accurate reference content.

[0170] In one possible embodiment, step S305 described above may include:

[0171] Step D1: Based on at least one evaluation metric of the candidate image, determine the score value of at least one scoring dimension of the candidate image, and based on the score value of at least one scoring dimension of the candidate image, analyze the quality score value of the imaging quality of the content to be identified in the candidate image.

[0172] Wherein, at least one evaluation metric for the candidate image includes: the confidence level in the recognition result of the candidate image, at least one of the at least one image feature and the target auxiliary evaluation metric;

[0173] Step D2: Select candidate images from the at least one candidate image according to the quality score of the candidate image to obtain the image filtering result.

[0174] For the candidate image, given the target auxiliary evaluation index of the candidate image, the confidence level in the recognition result of the candidate image, at least one image feature of the candidate image, and at least one of the target auxiliary evaluation index can be used as at least one evaluation index of the candidate image. Furthermore, based on the at least one evaluation index of the candidate image, a score value for at least one scoring dimension of the candidate image is determined, and based on the score value of the at least one scoring dimension of the candidate image, a quality score value for the imaging quality of the content to be identified in the candidate image is analyzed. Then, after obtaining each quality score value for each candidate image, candidate images can be selected from the at least one candidate image according to the determined quality score values ​​to obtain image filtering results. For example, the candidate image with the highest quality score value can be selected as the image filtering result, or candidate images with quality scores higher than a first score threshold and each scoring dimension higher than a second score threshold can be selected as the image filtering result.

[0175] It is understandable that each scoring dimension can uniquely correspond to one evaluation metric, or it can correspond to at least one evaluation metric, both of which are reasonable. Furthermore, there are various ways to analyze the imaging quality score of the content to be identified in the candidate image based on the scoring values ​​of at least one scoring dimension of the candidate image. For example, the scoring values ​​of at least one scoring dimension of the candidate image can be weighted to obtain the imaging quality score of the content to be identified in the candidate image.

[0176] There are multiple ways to determine the score value of at least one scoring dimension of the candidate image based on at least one evaluation index. For clarity and layout, other embodiments are described below.

[0177] In this embodiment, based on the confidence level in the recognition results of the candidate images, the at least one image feature, and the target auxiliary evaluation index, at least one evaluation index for the candidate images is set. Therefore, based on the at least one evaluation index, the candidate images are scored from at least one scoring dimension. Based on the score values ​​of the at least one scoring dimension of the candidate images, the quality score value of the imaging quality of the content to be identified in the candidate images is analyzed. Finally, candidate images are selected from at least one candidate image according to the quality score value to obtain the image filtering result. It can be seen that in this embodiment, scoring the candidate images from at least one scoring dimension makes the scoring of the candidate images more comprehensive. Therefore, by combining the score values ​​of at least one scoring dimension, the quality score value of the imaging quality of the content to be identified in the candidate images can be analyzed for image filtering, thereby improving the accuracy of image filtering.

[0178] In one possible embodiment, determining the score value of at least one scoring dimension of the candidate image based on at least one evaluation metric of the candidate image may include:

[0179] According to the scoring strategy corresponding to the evaluation index of the candidate image, the score value of the scoring dimension corresponding to the evaluation index of the candidate image is determined; wherein, different evaluation indexes correspond to different scoring dimensions.

[0180] In this embodiment, the scoring dimension and the evaluation index have a unique correspondence, and each evaluation index has its own scoring strategy; thus, for each evaluation index of the candidate image, the scoring value of the scoring dimension corresponding to the evaluation index can be determined according to the scoring strategy corresponding to the evaluation index, so as to score the imaging quality of the content to be identified in the candidate image.

[0181] For example, for each evaluation metric belonging to image features, if the larger the value of the evaluation metric, the better the imaging quality of the content to be identified, then the scoring strategy set for the evaluation metric may include a strategy where the metric value and the score value are positively correlated. If the smaller the value of the evaluation metric, the worse the imaging quality of the content to be identified, then the scoring strategy set for the evaluation metric may include a strategy where the metric value and the score value are negatively correlated. For evaluation metrics belonging to confidence, the scoring strategy set for the evaluation metric may include a strategy where the confidence level and the score value are positively correlated. For a first auxiliary evaluation metric that serves as an evaluation metric, the scoring strategy corresponding to the first auxiliary evaluation metric may include a strategy where the metric value and the score value are positively correlated.

[0182] It should be emphasized that the specific content of the scoring strategy set for the evaluation indicators in this application embodiment is not specifically limited.

[0183] In this embodiment, the evaluation indicators and scoring dimensions have a unique correspondence, and each evaluation indicator has its own scoring strategy, which can ensure the comprehensiveness of the scoring dimensions and the accuracy of the scoring for each scoring dimension. This makes the scoring of each candidate image more comprehensive and accurate. Thus, by combining the scoring values ​​of at least one scoring dimension, the quality score of the character imaging quality of the candidate image can be analyzed for image screening, which can improve the accuracy of image screening.

[0184] In one possible embodiment, determining the score value of at least one scoring dimension of the candidate image based on at least one evaluation metric of the candidate image may include:

[0185] For the candidate image, based on the scoring strategy corresponding to the evaluation index combination, determine the score value of a scoring dimension corresponding to the evaluation index combination.

[0186] For the candidate image, excluding the combination of evaluation indicators, a score value for a scoring dimension corresponding to the evaluation indicator is determined according to the scoring strategy corresponding to the evaluation indicator.

[0187] In this embodiment, for at least one evaluation index of the candidate image, some of the evaluation indexes can be divided into an evaluation index combination, which uniquely corresponds to a scoring dimension; each evaluation index other than the evaluation index combination can have a unique correspondence with a scoring dimension.

[0188] It is understood that evaluation indicators belonging to the same evaluation indicator combination can be evaluation indicators of the same category. For example, at least one image feature, as an evaluation indicator of the same category, can be classified into one evaluation indicator combination. Alternatively, evaluation indicators belonging to the same evaluation indicator combination can be related evaluation indicators (for example, the level of one evaluation indicator affects the credibility of another evaluation indicator's assessment of character imaging quality). For example, the confidence level in the recognition result is related to the second auxiliary evaluation indicator, or the confidence level in the recognition result is related to the first auxiliary evaluation indicator. This application embodiment does not specifically limit the conditions that evaluation indicators in the same evaluation indicator combination must meet; in specific applications, these conditions can be set according to actual circumstances.

[0189] The scoring strategies for individual evaluation indicators can be found in the relevant descriptions of the above embodiments, and will not be repeated here.

[0190] The scoring strategy for a combination of evaluation indicators can include various methods. Optionally, in one implementation, the scoring strategy for the combination of evaluation indicators can include:

[0191] Based on the target indicators in the evaluation indicator combination, an initial score is obtained;

[0192] The scoring correction coefficient is determined based on the evaluation indicators other than the target indicator in the evaluation indicator combination.

[0193] The initial score value is corrected using the score correction coefficient to obtain the score value of one score dimension corresponding to the evaluation index combination.

[0194] In this system, the target indicator can be the primary indicator in the evaluation indicator combination, while other evaluation indicators excluding the target indicator can be used as secondary indicators. An initial score can be obtained based on the scoring strategy corresponding to the target indicator. Then, based on the other evaluation indicators in the combination excluding the target indicator, a score correction coefficient is determined according to the mapping relationship between the indicator values ​​and correction coefficients. Using the score correction coefficient, the initial score value is corrected to obtain the score value for one scoring dimension of the evaluation indicator combination. This implementation method makes the score value corresponding to the evaluation indicator combination more accurate.

[0195] In other implementation methods, the scoring strategies corresponding to the combination of evaluation indicators can also include:

[0196] For each evaluation indicator in the evaluation indicator combination, an initial score is made according to the scoring strategy corresponding to that evaluation indicator to obtain the initial score value for each evaluation indicator.

[0197] The initial scores are weighted and calculated to obtain the score values ​​for the corresponding scoring dimensions of the evaluation index combination.

[0198] It should be emphasized that the above description of the implementation method of the scoring strategy corresponding to the combination of evaluation indicators is merely an example and should not be construed as limiting the embodiments of this application.

[0199] In this embodiment, evaluation indicators with a relationship (such as evaluation indicators of the same type or evaluation indicators with a correlation) are grouped into the same evaluation indicator combination, and the evaluation indicator combination uniquely corresponds to a scoring dimension, so that the setting of the scoring dimension is more in line with the characteristics of each evaluation indicator; thereby, by combining the scoring value of at least one scoring dimension, the quality score value of the imaging quality of the content to be identified in the candidate image is analyzed for image screening, which can improve the accuracy of image screening.

[0200] In addition, embodiments of this application also provide an image filtering method, such as... Figure 4 As shown, the image filtering method may include the following steps:

[0201] S401, Obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0202] S402, perform image recognition processing on the reference image of the candidate image to obtain the reference content of the candidate image;

[0203] Wherein, the content to be identified (such as character content or target object) in the reference image of the candidate image is the same as the content to be identified in the candidate image;

[0204] S403, Based on the reference content of the candidate image, determine the target auxiliary evaluation index of the candidate image;

[0205] S404, based on the recognition result of the candidate image and the target auxiliary evaluation index, determine the image screening result from the at least one candidate image.

[0206] Optionally, in one implementation, the target auxiliary evaluation index includes a first auxiliary evaluation index and / or a second auxiliary evaluation index; the first auxiliary evaluation index characterizes the degree of consistency between the result content represented by the recognition result of the candidate image and the reference content; the second auxiliary evaluation index characterizes the confidence level of the reference content of the candidate image.

[0207] Optionally, in one implementation, the reference image for the candidate image is obtained by:

[0208] The candidate image is subjected to image transformation processing, and a reference image is determined based on the image obtained after image transformation processing.

[0209] Optionally, in one implementation, performing image recognition processing on a reference image of the candidate image to obtain reference content of the candidate image may include:

[0210] If no true value exists for the candidate image, perform image recognition processing on the reference image of the candidate image to obtain the reference content of the candidate image;

[0211] The method may further include:

[0212] If the true value exists, the true value is determined as the reference content of the candidate image, and the step of determining the target auxiliary evaluation index of the candidate image based on the reference content of the candidate image is triggered.

[0213] Optionally, in one implementation, determining the image selection result from the at least one candidate image based on the recognition result of the candidate image and the target auxiliary evaluation index includes:

[0214] Based on at least one evaluation metric of the candidate image, a score value for at least one scoring dimension of the candidate image is determined, and based on the score value for at least one scoring dimension of the candidate image, a quality score value for the imaging quality of the content to be identified in the candidate image is analyzed; wherein, the at least one evaluation metric of the candidate image includes: the confidence level in the recognition result of the candidate image and at least one of the target auxiliary evaluation metrics.

[0215] Based on the quality score of the candidate image, a candidate image is selected from the at least one candidate image to obtain the image filtering result.

[0216] The execution subject, application scenarios, various concepts, and exemplary implementation methods of each step involved in the image filtering method provided in this embodiment can be found in the corresponding content of the foregoing embodiments, and will not be repeated here.

[0217] In another image filtering method provided in this application embodiment, for at least one candidate image, a recognition result obtained by image recognition processing is obtained. Furthermore, image recognition processing is performed on a reference image of the candidate image to obtain reference content of the candidate image. Based on the reference content of the candidate image, a target auxiliary evaluation index for the candidate image is determined. Finally, based on the recognition result and the target auxiliary evaluation index of the candidate image, an image filtering result is determined from the at least one candidate image. It can be seen that this scheme performs image filtering based on the recognition result of image recognition processing, combined with the target auxiliary evaluation index determined based on the reference content in the reference image. This eliminates the need to rely on the ground truth of the candidate image, enabling diverse filtering criteria. Compared to existing technologies, this reduces the impact of the recognition effect of image recognition processing on image filtering, thereby improving the accuracy of image filtering.

[0218] Considering that the above-described image selection method can be applied to the imaging recommendation process, based on the above-described image selection method, this application embodiment also provides an imaging recommendation method, such as... Figure 5 As shown, the imaging recommendation method may include the following steps:

[0219] S501, obtain at least one candidate image for imaging recommendation; wherein, the at least one candidate image is: an image obtained for the same target object containing character content based on different imaging configuration information;

[0220] S502, according to the image filtering method provided in the embodiments of this application, the image filtering result is determined based on the at least one candidate image;

[0221] S503, based on the imaging configuration information when the image screening result was captured, determine the imaging recommendation result.

[0222] The aforementioned imaging recommendation can be either offline or online. This application does not specifically limit the implementation of S501. Any method that can obtain at least one candidate image for imaging recommendation can be applied to the embodiments of this application.

[0223] Furthermore, the imaging configuration information may include one or more of the following information that can affect the imaging effect: shooting parameters, imaging algorithm, lighting conditions, etc.; at least one of the information in different imaging configuration information is different.

[0224] The specific implementation of step S502 can be described in the relevant content introduced in the above embodiments of the image filtering method, and will not be repeated here.

[0225] For example, the imaging recommendation is specifically an offline imaging recommendation, and determining the imaging recommendation result based on the imaging configuration information at the time of shooting according to the image selection result may include:

[0226] Construct the association between the imaging configuration information of the image selection result when it is captured and the object type of the target object to obtain the imaging recommendation result;

[0227] For example, the imaging recommendation is specifically an online imaging recommendation, and determining the imaging recommendation result based on the imaging configuration information at the time of shooting according to the image selection result may include:

[0228] One or more of the following information are used as the imaging recommendation result: the image filtering result, the recognition result of the image filtering result, and the imaging configuration information when the image filtering result was captured.

[0229] The imaging recommendation method provided in this application, after obtaining at least one candidate image for imaging recommendation, can determine the image filtering result based on the at least one candidate image according to the image filtering method provided in this application; and, based on the imaging configuration information at the time the image filtering result was captured, determine the imaging recommendation result. It is evident that, in the imaging recommendation process, the image filtering of at least one candidate image can reduce the impact of image recognition processing on image filtering compared to existing technologies, ultimately reducing the impact of image recognition processing on imaging recommendation, thereby improving the accuracy of imaging recommendation.

[0230] To better understand the imaging recommendation method provided in this application, a specific embodiment is described below. In this specific embodiment, the content to be identified is the character content contained in the target object, and the image recognition processing is specifically text recognition.

[0231] The technical solution provided in this specific embodiment introduces image feature extraction and evaluation methods other than text recognition results. At the same time, it evaluates the character imaging quality of candidate images in online recipe recommendation without truth values ​​by using reference text (i.e., reference content determined by reference images). Thus, it realizes multi-dimensional comprehensive evaluation in the imaging recommendation process, ensuring that the imaging recommendation result is the best imaging recommendation result recognized by both automatic recognition and human eye recognition.

[0232] The imaging recommendation method provided in this specific embodiment may include steps 1-8, wherein steps 1-7 are the single-image evaluation stage (i.e., the stage of scoring the quality of each candidate image), and step 8 is the result recommendation stage. The specific steps are as follows:

[0233] Step 1: Obtain the image from which character recognition is required. Other supporting information;

[0234] Other auxiliary information may be null, or may include one or more of the target additional information and string truth value S (i.e., the text truth value mentioned above);

[0235] The additional information for this target may include character recognition region information R (i.e., the additional information mentioned above used to characterize size information), character shape, etc.

[0236] And, images This refers to any one of the candidate images, which can form a candidate image queue. =, the total number of candidate images is N+1, This represents any image in the candidate image queue.

[0237] Step 2, processing the image based on deep learning algorithms. Perform text recognition to obtain images The recognition result of the image The character content in the recognition result is denoted as The confidence level is denoted as .

[0238] Step 3: Obtain the reference string (i.e., the aforementioned reference content) based on different input conditions:

[0239] The different input conditions specifically refer to: whether the string truth value S exists in other auxiliary information; specifically:

[0240] In the absence of a string truth value S, the image is transformed according to the defined image transformation method. Perform image transformation processing to generate an image. Corresponding reference image ; and based on deep learning algorithms, the reference image Perform text recognition to obtain a reference image. The recognition results, reference image The character content in the recognition result is used as a reference string and denoted as . and reference image The confidence score in the recognition result is used as the confidence score of the reference string and denoted as . .

[0241] Image transformation processing may include, but is not limited to, one or more of the following: grayscale transformation processing, pixel transformation processing, and geometric transformation processing.

[0242] If a string truth value exists, there is no need to modify the image. For image transformation processing, the true value of the string is directly used as the reference string and denoted as... And the confidence level is recorded as And the confidence level is .

[0243] The confidence level of the reference string is the second auxiliary evaluation index in the above embodiments.

[0244] Step 4: Compare the reference string and the image. The character content in the recognition results is recorded, and the consistency evaluation between the two is recorded. (i.e., similarity, corresponding to the first auxiliary evaluation index mentioned above);

[0245] Among them, consistency evaluation is used for reference strings and images. The consistency of character content in the recognition results can be calculated in various ways, including but not limited to: calculation methods based on indicators such as edit distance, number of common characters, and total length of common substrings. Different indicators have different tolerances for differences between strings. In specific applications, the specific calculation method can be selected according to the actual situation.

[0246] Step 5, extract the image At least one image feature (focusing only on the character recognition region), and the image feature is denoted as... ;

[0247] The at least one image feature may include, but is not limited to, one or more of the following: average gray level, gray level variance, contrast, average gradient, gray level moment, etc., or features extracted by a feature extraction model; and, when performing feature extraction, it can be performed by a pre-trained deep learning model.

[0248] In addition, when there is additional target information in other auxiliary information, the type of image features to be extracted and / or the feature extraction method used can be determined based on the additional target information.

[0249] Step 6: After processing each candidate image using steps 1-5, each candidate image in the candidate image queue can be obtained. The set of evaluation indicators, which includes: , , and ;in, Representing candidate images The confidence level in the recognition results. Candidate images The confidence level of the reference string; Candidate images Consistency evaluation, Candidate images At least one image feature.

[0250] Step 7: Perform fusion processing on the various evaluation indicators to calculate the comprehensive score (i.e., the quality score value of each candidate image mentioned above):

[0251] This fusion process can be represented by the following formula:

[0252] );

[0253] in, Candidate images Quality score, () is the processing function for implementing fusion processing.

[0254] The specific fusion processing method is as described above: based on the various evaluation indicators of the candidate image, determine the score values ​​of multiple scoring dimensions of the candidate image, and based on the score values ​​of multiple scoring dimensions of the candidate image, analyze the quality score value of the character imaging quality (corresponding to the imaging quality of the content to be identified mentioned above).

[0255] Step 8: Maintain an evaluation index queue (i.e., a queue of quality scores) corresponding to the candidate image queue, and output the imaging configuration information of the candidate image with the highest quality score when it was captured as the imaging recommendation result.

[0256] Understandably, the candidate image with the highest quality score, along with the imaging configuration information at the time of capture, can also be used as the imaging recommendation result for output.

[0257] The candidate image with the highest quality score is selected using the following formula:

[0258] ;

[0259] in, The Index() function represents the index function and serves as the image identifier for the candidate image with the highest quality score.

[0260] The characterization index is used to find the candidate image with the highest quality score.

[0261] In addition, to better understand the imaging recommendation method implemented in steps 1-8 above, Figure 6 An exemplary flowchart is provided, illustrating the implementation process of this imaging recommendation method:

[0262] S601, acquire images and other auxiliary information;

[0263] S601 corresponds to step 1 above;

[0264] S602, Original image text recognition, to obtain the original image recognition result and confidence level;

[0265] Does S603 contain a true value?

[0266] That is, determine whether there is a text truth value. If yes, directly use the text truth value as the reference string and determine the confidence level to obtain the reference string and confidence level; if no, trigger S604-S605 to obtain the reference string and confidence level.

[0267] S604, Image Transformation;

[0268] S605, Transformed Image Text Recognition;

[0269] S604-S605 correspond to the processing steps in step 3 above for the case where the string truth value S does not exist.

[0270] S606, Other feature extraction to obtain other features;

[0271] S606 corresponds to step 5 above.

[0272] S607, Feature Fusion;

[0273] S607 corresponds to the processing steps 4 and 6 and 7 described above.

[0274] S608, Generate a candidate sample queue;

[0275] This candidate sample queue is the same as the evaluation index queue mentioned above.

[0276] S609, recommended result;

[0277] S610, optimal imaging and corresponding configuration;

[0278] S608-S610 correspond to step 8 above and belong to the result recommendation stage, while S602-S607 belong to the single image evaluation stage; among them, the best image is the candidate image with the highest quality score, and the corresponding configuration is the imaging configuration information when the candidate image with the highest quality score is captured.

[0279] The imaging recommendation method provided in this specific embodiment introduces a reference string generation method based on image transformation, ensuring stable recognition of the original image under various conditions and improving the reliability of recommendation results in the absence of ground truth. Furthermore, by comprehensively considering the text recognition results obtained through deep learning and other image features within the ROI (Region of Interest) range (i.e., the text recognition region), it recommends effective optimal imaging and configuration that simultaneously meets human visual evaluation and OCR (Optical Character Recognition), reducing the configuration difficulty of the imaging recommendation process. Moreover, through scalable additional features and flexible recommendation logic (besides the logic for handling ground truth, it also allows for flexible generation of feature extraction strategies based on different target additional information), the success rate and versatility of the imaging recommendation method are improved.

[0280] Furthermore, in image recognition processing scenarios (also known as online recognition scenarios), related technologies typically require pre-setting imaging conditions for a certain type of object (e.g., industrial workpieces, paper carriers, electronic display interfaces, signs, etc.). Then, for the object to be recognized, imaging is performed based on the set imaging conditions to obtain the imaging result, and image recognition processing is then performed based on the imaging result. However, to ensure the accuracy of the object's image recognition results, a large number of imaging conditions are usually set. This means that during the recognition process, a large number of imaging conditions need to be polled (i.e., imaging is performed based on each imaging condition), resulting in low recognition efficiency due to the large number of imaging results. In particular, to handle mixed object detection scenarios (i.e., scenarios where the type of the object to be recognized is not fixed), each type may have its own large number of imaging conditions. Therefore, the number of imaging conditions that need to be polled during the recognition process is even greater, exacerbating the problem of low recognition efficiency.

[0281] Based on this, embodiments of this application provide a recognition method, offering a data foundation for improving recognition efficiency. It should be noted that this image recognition processing method can be a further method based on the image filtering method described above. The recognition method provided in this application is applied to a client-side application; such as... Figure 7 As shown, the method may include:

[0282] S701 receives configuration requests for samples;

[0283] S702, in response to the configuration request, display a first recognition result and a first evaluation index for the sample; the first evaluation index indicates the imaging quality of the corresponding first recognition result; the first recognition result is a recognition result obtained by performing image recognition processing on multiple imaging results of the sample, and the corresponding first evaluation index satisfies the first screening condition; the multiple imaging results are obtained by imaging the sample based on set multi-group imaging conditions.

[0284] S703, in response to user interaction, sets the imaging conditions corresponding to the first recognition result to the first imaging conditions for the sample to complete the configuration.

[0285] For example, the client can take the form of an app (mobile application), a web page (web page), or a computer program (desktop application); of course, the client can take other forms, such as command-line tools, smart devices, etc. This application does not limit the specific form of the client; in specific applications, it can be flexibly selected based on actual business scenarios and service requirements.

[0286] The sample can be an object of a certain type (such as an object containing character content), and the specific type can be set according to the actual imaging configuration requirements (i.e., imaging condition configuration requirements) and / or recognition requirements. For example, the sample can be an industrial workpiece, paper carrier, electronic display interface, signboard, or other carrier or object that carries text, numbers, symbols, and / or codes and has imaging configuration requirements. Furthermore, the number of samples can be one or more. For example, different samples can belong to different types of objects. In this case, S701 specifically refers to a configuration request for at least one sample, and the number of samples can be set according to the actual imaging configuration requirements and / or recognition requirements. It should be noted that the imaging conditions in this embodiment are the imaging configuration information in the above embodiments, and the condition content in the imaging conditions can include user-editable content items and / or user-uneditable content items; the imaging configuration requirements refer to configuring imaging conditions for the sample that enable the imaging quality to meet quality requirements (such as clear character imaging or clear image of the sample itself) for subsequent processing (such as online recognition as described below).

[0287] Furthermore, for example, the configuration request may be issued by the user clicking the reservation button. For example, after selecting a sample through the client, the user can click the start button, thereby the client receives the sample configuration request. For example, the configuration request may also be sent to the client through a device communicating with the client when the configuration triggering conditions are met. For example, the configuration triggering conditions include imaging the sample to obtain the multiple imaging results, or other conditions set according to actual business needs.

[0288] Upon receiving a configuration request, the client can display a first recognition result and a first evaluation index for the sample, providing the user with relevant information about the image recognition processing of the sample. This allows the user to intuitively understand the imaging quality under the candidate imaging conditions recommended by the client. The number of displayed first recognition results can be one or more; correspondingly, since the first recognition result corresponds to the first evaluation index, there can also be one or more first evaluation indices. Furthermore, for example, the specific value of the first evaluation index can be a score, a grade, etc.; any index capable of distinguishing between high and low imaging quality is acceptable.

[0289] Of course, in other implementations, in order to enable users to have a more comprehensive understanding of the imaging quality of the sample to be identified under the recommended candidate imaging conditions, in response to the configuration request, the client may also display the imaging result to which the first identification result belongs, and / or, imaging information used to characterize the imaging conditions corresponding to the first identification result, wherein the displayed imaging information may include, but is not limited to: the identification information and / or condition content of the imaging conditions corresponding to the first identification result.

[0290] Furthermore, it is understood that before displaying the first recognition result and the first evaluation index, the client can obtain the first recognition result and the first evaluation index of the sample and display them through an interface or pop-up window. The first recognition result and the first evaluation index of the sample can be determined by the client itself, or determined by the client's backend server and fed back to the client. For example, the determination of the first recognition result and the first evaluation index of the sample can include: obtaining multiple imaging results of the sample; performing image recognition processing on each of the multiple imaging results to obtain a recognition result corresponding to each imaging result (such as a text recognition result, or a recognition result related to the sample itself); evaluating the imaging quality of the recognition result corresponding to each imaging result through a set imaging quality evaluation method to obtain a first evaluation index corresponding to each recognition result; selecting recognition results whose corresponding first evaluation index meets the first screening condition from the obtained recognition results to obtain the first recognition result; and determining the first evaluation index of the first recognition result after obtaining the first recognition result.

[0291] The multi-group imaging conditions are pre-set imaging conditions, which can be stored locally on the client or on the corresponding backend server. Furthermore, these multi-group imaging conditions can be set by the user or generated by the client or backend server according to a predetermined method; both are acceptable. Additionally, it should be noted that the multi-group imaging conditions can be shared by different types of samples or used exclusively for a single sample.

[0292] For example, the first screening condition may include: the highest indicator value, the indicator value being higher than a predetermined threshold, or being ranked in the top N positions (the value of N can be set according to actual conditions), or other conditions set according to actual business needs. Furthermore, for example, the method of obtaining the imaging results of the sample may include, but is not limited to: obtaining multiple imaging results obtained from pre-imaging (e.g., downloading from a given download address), or, according to multiple imaging conditions, acquiring images of the sample through an image acquisition device to obtain multiple imaging results; both are reasonable.

[0293] It should be noted that the imaging result of the sample can be understood as the candidate image in the image screening method described above. Accordingly, the output first recognition result and first evaluation index can be considered as: the recognition result and quality score of the image screening result (i.e., the screened image) obtained through the image screening method (or, of course, information such as the rating level formed based on the quality score). In other words, the image screening method of the above embodiment can be applied to the process of evaluating the imaging quality of the recognition result corresponding to each imaging result in this embodiment.

[0294] After displaying the first identification result and the first evaluation index for the sample, the user can analyze whether they approve of the current imaging quality based on the displayed information and interact with the client. The client then responds to the user's interaction by setting the imaging file corresponding to the first identification result to the first imaging conditions for the sample to complete the configuration. The imaging file corresponding to the first identification result refers to the imaging conditions used in the imaging result with the first identification result.

[0295] Optionally, the step of responding to user interaction and setting the imaging condition corresponding to the first recognition result to a first imaging condition for the sample to complete the configuration may include:

[0296] In response to the interactive instruction that confirms the characterization, the imaging conditions corresponding to the first identification result are set to the first imaging conditions for the sample to complete the configuration;

[0297] In response to an interactive instruction to reset the representation, the first identification result for the sample and the first evaluation index corresponding to the first identification result are displayed again.

[0298] For example, the interactive instruction to confirm can be issued by performing a first predetermined operation; for example, a first button element to confirm can be displayed on the interface that displays the first identification result and the second evaluation index, and performing the first predetermined operation is to click the first button element; or, for example, a first device button is stored on the device to which the client belongs, and performing the first predetermined operation is to click the first device button.

[0299] For example, the interaction instruction representing a reset can be issued by performing a second predetermined operation; for example, a second button element representing a reset can be displayed on the interface displaying the first identification result and the second evaluation index, and performing the second predetermined operation is to click the second button element; or, for example, there is a second device button on the device to which the client belongs, and performing the second predetermined operation is to click the second device button.

[0300] It is understood that before re-displaying the first identification result and the first evaluation index corresponding to the first identification result for the sample, the generation process of the first identification result and the first evaluation index for the sample can be re-executed, so that the client can obtain the regenerated first identification result and the first evaluation index and display them; and after re-displaying, S703 can continue to be executed.

[0301] After completing the first imaging condition for the sample, at least one first imaging condition that meets the quality requirements is selected from multiple imaging conditions for the sample, thus obtaining the "imaging formula" of the sample.

[0302] To facilitate understanding of the implementation process of S701-S703, text recognition will be used as an example below. Figure 9 An example will be provided. For example... Figure 9 As shown, after the client displays the optimization parameter tuning interface (i.e., the interface for setting imaging conditions) corresponding to the sample, the user issues a configuration request for the sample by clicking the "Start" button. In response to the configuration request, the client can display "00007" (i.e., display the first recognition result) in the input area corresponding to the String field and "193" (i.e., display the first evaluation index) in the input area corresponding to the Score field. Furthermore, after the user issues a confirmation instruction via the "Accept" button, the client can respond to the user's instruction by setting the imaging condition corresponding to "00007" to the first imaging condition for the sample to complete the configuration; or, after the user issues a reset instruction via the "Reset" button, the client can re-display information in the input areas corresponding to the String field and the Score field until the user issues a confirmation instruction to complete the configuration. Figure 9 The content of Interface 1, Interface 2, and other information areas can be set according to the actual situation.

[0303] In this embodiment, the client can receive a configuration request for a sample, and in response to the configuration request, display the first recognition result and the first evaluation index for the sample; and, in response to user interaction, set the imaging conditions corresponding to the first recognition result as the first imaging conditions for the sample to complete the configuration. It is evident that in this solution, through simple interaction between the user and the client, a configuration process for the first imaging conditions of the sample is added, thereby enabling the online recognition process for objects of the same type as the sample to reduce the number of imaging conditions required while ensuring recognition accuracy, providing a data foundation for improving recognition efficiency; furthermore, through simple interaction between the user and the client, the configuration of imaging conditions for the sample can be completed, thereby lowering the operational threshold, simplifying the configuration process, and improving configuration efficiency and user experience.

[0304] In one possible embodiment, in Figure 7 Based on the embodiments, such as Figure 8 As shown, the identification method may further include:

[0305] S704, Receive an identification request for the object to be identified;

[0306] S705, in response to the recognition request, a second recognition result and a second evaluation index for the object are displayed; the second evaluation index indicates the imaging quality of the corresponding second recognition result; the second recognition result is a recognition result obtained by performing image recognition processing on at least one imaging result of the object, and the corresponding second evaluation index satisfies the second screening condition; the at least one imaging result is obtained by imaging the object based on at least one of the first imaging conditions.

[0307] After configuring the first imaging conditions for at least one sample, these conditions can be applied to an object recognition scenario. For example, this scenario could be a mixed object detection scenario, where the type of the object to be identified is not fixed, but belongs to one of the types involved in the at least one sample. In a mixed object detection scenario, compared to each sample corresponding to multiple imaging conditions, since at least one sample corresponds to a small number of first imaging conditions, the candidate range of imaging conditions that can guarantee imaging quality can be greatly narrowed. This reduces the number of imaging results for the object to be detected, ultimately improving the recognition efficiency of the object while maintaining the accuracy of the recognition results.

[0308] Specifically, the client can receive an identification request for an object to be identified, and in response to the identification request, display a second identification result and a second evaluation index for the object. For example, the identification request can be issued by the user interacting with the client; alternatively, it can be sent to the client by an electronic device communicating with the client when the object to be detected meets the identification triggering conditions. For instance, the electronic device senses that the object to be detected has been moved to a predetermined identification area and sends an identification request to the client.

[0309] It is understood that the client can obtain the second identification result and second evaluation index of the object before displaying the second identification result and second evaluation index of the object. The generation process and generator (such as the client itself or the backend server) of the second identification result and second evaluation index of the object are similar to those of the first identification result and first evaluation index of the aforementioned sample. The difference lies in that the imaging result of the object is obtained by imaging based on at least one first imaging condition, and the second filtering condition can be the same as or different from the first filtering condition. The imaging result of the object can be obtained after receiving the identification request or before receiving the identification request; both are reasonable. For example, the second filtering condition can include the highest index value, an index value higher than a predetermined threshold, being ranked in the top N positions (the value of N can be set according to actual conditions), or other conditions set according to actual business needs.

[0310] Optionally, in one implementation, the identification method may further include:

[0311] Display the imaging result to which the second identification result belongs, and / or, imaging information used to characterize the first imaging condition corresponding to the second identification result.

[0312] Wherein, the first imaging condition corresponding to the second identification result refers to the first imaging condition utilized by the imaging result to which the second identification result belongs. Furthermore, for example, the displayed imaging information may be the condition identifier and / or condition content of the first imaging condition corresponding to the second identification result.

[0313] Therefore, by displaying the imaging results and / or imaging information to which the second recognition result belongs, users can intuitively view richer information about the second recognition result and gain a comprehensive understanding of the relevant information regarding the image recognition processing of the object.

[0314] In this embodiment, the first imaging condition of the at least one sample is applied to a mixed object detection scenario, which greatly narrows down the candidate range of imaging conditions that can guarantee imaging quality to a certain extent. Furthermore, after receiving a recognition request for the object to be identified, the client can respond to the recognition request by displaying a second recognition result and a second evaluation index for the object. Therefore, this embodiment can balance the accuracy and efficiency of object image recognition processing.

[0315] In one possible embodiment, before receiving the configuration request for the sample, the identification method may further include:

[0316] In response to input configuration information, associate the configuration information with the sample;

[0317] The configuration information is used for the image recognition processing and / or the generation of the first evaluation index.

[0318] For example, the configuration information refers to the content items in the imaging conditions that are configurable by the user; in other words, for the configurable content items in the imaging conditions that are open to the user, the user can customize the settings according to actual needs to better adapt to the sample configuration process; correspondingly, associating the configuration information with the sample can be understood as updating the configuration information to the first imaging conditions of the sample. Taking text recognition as an example, the configuration information may include, but is not limited to: information representing the morphological features of characters (such as font, font size, etc.), the verification method of the recognition result, the text truth value (such as the text truth value of the sample), the text region (such as the region for complete text recognition, or a region for a single character), etc.

[0319] And, in one implementation, in response to input configuration information, associating the sample with the configuration information includes:

[0320] Display the configuration interface;

[0321] In response to the input of configuration information on the configuration interface, the setting information is associated with the sample.

[0322] In this implementation, a configuration information input interface can be provided to the user through the configuration interface, ensuring that the configuration process is intuitive and visual.

[0323] For example, in other implementations, associating the configuration information with the sample in response to input configuration information may include:

[0324] In response to receiving a configuration file containing configuration information, the configuration file is parsed to obtain the configuration information, and the configuration information is associated with the sample.

[0325] In this implementation, the configuration information is carried in a configuration file. The client receives the configuration file, parses it to obtain the configuration information, and associates it with the sample. It is understood that carrying the configuration information in the configuration file decouples the parameters from the client program, facilitating independent modification, distribution, persistent storage, and cross-device reuse of the configuration, significantly reducing the workload of configuring multi-sample imaging conditions.

[0326] Optionally, in one implementation, associating the sample with the configuration information in response to input configuration information may include:

[0327] A reference image of the sample is shown;

[0328] In response to an editing operation on the reference image, the edited content corresponding to the editing operation is displayed on the reference image; wherein the editing operation includes at least one of the following:

[0329] Recognize and draw the location of the region (e.g., draw the text region); adjust the location of the region (e.g., adjust the location of the text region); adjust the size of the region (e.g., adjust the size of the text region).

[0330] In this implementation, after receiving a loading instruction for the reference image of the sample, the client can load and display the reference image of the sample. Based on the displayed reference image, the user can enter configuration information through image editing, which improves the convenience of configuration operations and optimizes the human-computer interaction experience.

[0331] Furthermore, in response to the editing operation on the reference image, the edited content corresponding to the editing operation is displayed on the reference image in real time; wherein, the recognition area indicated by the editing operation can be an operation on a complete recognition area (such as a text area) or an operation on a local area (such as a single character area in a text area), that is, the ROI (Region of Interest) can be a complete recognition area or a local area; and the drawing and adjustment operation indicated by the editing operation is the drawing and adjustment of the ROI box.

[0332] It is understood that the reference image may be displayed in a partial area of ​​the aforementioned configuration interface, or on other interfaces distinct from the configuration interface (e.g., related interfaces accessed from the configuration interface), or in a pop-up area; this application does not limit the display medium of the reference image. For example, Figure 10A schematic diagram of a configuration interface is provided. The configuration interface displays an image editing area for displaying a reference image 1001, and a field configuration area for inputting various configuration information (e.g., character font, text truth value, etc.). Field 1, Field 2, and Field 3 are the field names of configuration information that the user can input. The user can perform editing operations on the reference image 1001 in the image editing area, such as drawing a ROI box 1002.

[0333] It is understood that after editing is completed (e.g., when the user clicks the save button), the region information of the recognition area indicated by the current edited content can be associated with the sample for the image recognition processing and / or the generation of the first evaluation index.

[0334] Optionally, in one implementation, the client provides a truth value editing area for inputting text truth values, and this is a first configuration item that can be omitted.

[0335] In this implementation, the truth value editing area can be displayed in a local area of ​​the aforementioned configuration interface, or in other interfaces different from the configuration interface (e.g., related interfaces accessed from the configuration interface), or in a dedicated pop-up area; this application does not limit the display medium of the truth value editing area.

[0336] Furthermore, the truth value editing area is a first configuration item that can be omitted. Specifically, the user can choose to configure a truth value or not (for example, by setting the truth value editing area to empty or entering default content representing an unconfigured state). It is understood that the truth value is an optional configuration item. When a truth value is configured, quality assessment can be performed according to the processing mechanism corresponding to the presence of a truth value (such as the processing mechanism in the image filtering method described above); while when no truth value is configured, quality assessment can be performed according to the processing mechanism corresponding to the absence of a truth value (such as the processing mechanism in the image filtering method described above).

[0337] Optionally, in one implementation, the client provides a second configuration item for the location function of the content to be identified, the second configuration item having a user-switchable option to enable or disable it.

[0338] In this implementation, the client provides the user with a second configuration item, which the user can operate to enable or disable. It is understood that when the positioning function is enabled, for the imaging result of the sample (or the imaging result of the object to be detected), the recognition area of ​​the imaging result can be located at least through the positioning function, which can be used for the generation of recognition results and / or the generation of evaluation indicators (of course, the user can also input the recognition area while the positioning function is enabled, and the two methods can be combined to determine the final recognition area). When the positioning function is not enabled, the user can input the location information of the recognition area through the aforementioned input configuration information. It is understandable that the positioning function can be enabled by default during the offline configuration stage (i.e., the recognition process of S701-S703 mentioned above); while the positioning function can be disabled by default during the online recognition stage (i.e., the recognition process of S704-S705 mentioned above). At this time, the location information of the recognition area involved in the first imaging condition can be used for image recognition processing and / or evaluation index generation. Furthermore, the recognition error rate when using the location information in the first imaging condition can be statistically analyzed. If the error rate exceeds a predetermined threshold, the positioning function can be enabled.

[0339] The client provides a second configuration option, making the recognition method of the recognition area more flexible and diverse, allowing users to choose according to their needs.

[0340] In this embodiment, in response to input configuration information, the configuration information is associated with the sample; the configuration information is used for the image recognition processing and / or the generation of the first evaluation index. Thus, users can independently set the configuration information for the recognition process of the adapted sample according to actual needs, enabling different samples to call matching recognition parameters for detection, effectively improving the adaptability and flexibility of the recognition samples.

[0341] To facilitate understanding of the recognition method provided in the embodiments of this application, text recognition will be used as an example below. Figure 11 The recognition principle is illustrated by an example. For example... Figure 11 The identification method may include two stages: offline configuration (corresponding to S701-S703 above) and online identification (corresponding to S704-S705 above).

[0342] The offline configuration process involves three samples: slice type A, slice type B, and slice type C, which are three types of workpieces. Each slice type contains character content. Each slice type can correspond to multiple image conditions. For example, slice type A corresponds to recipe group a (such as recipe a_1, recipe a_2, and recipe a_3), slice type B corresponds to recipe group b (such as recipe b_1, recipe b_2, and recipe b_3), and slice type C corresponds to recipe group c (such as recipe c_1, recipe c_2, and recipe c_3). Furthermore, ROI can be drawn based on the reference image (which displays character content) to obtain the text area input by the user.

[0343] The client can receive configuration requests for the three samples, obtain and display the first recognition result and first evaluation index for each sample, and respond to user interaction by setting the imaging conditions corresponding to the first recognition result as the first imaging conditions for the sample to complete the configuration. Specifically, formula a_1 is set as the formula corresponding to slice A (i.e., the first imaging condition for slice A), formula b_2 is set as the formula corresponding to slice B (i.e., the first imaging condition for slice B), and formula c_1 is set as the formula corresponding to slice C (i.e., the first imaging condition for slice C). Furthermore, the generation process of the first text result and first evaluation index for each sample includes: obtaining multiple imaging results for each sample, and then performing quality evaluation processing through light source polling (i.e., performing quality evaluation on the multiple imaging results of the sample through a quality evaluation method). This light source polling process includes text localization functionality, which is enabled by default, and a polling recommendation process, which can be considered as formula recommendation. In addition, each formula may include, but is not limited to: the recognition ROI (i.e., the location information of the text region), the light source position, the model ID (such as the model ID of the text recognition model or the ID of the quality evaluation model), and the light source intensity.

[0344] During online recognition, a formula for slice type N is applied for object recognition, where N... Specifically, the client receives an identification request for an object to be identified; in response to the identification request, it obtains and displays at least a second identification result and a second evaluation index for the object (corresponding to...). Figure 11 The process of generating the second identification result and the second evaluation index for the object can be considered as a recipe polling process (involving text localization (default off) and polling recommendation), which may include: text recognition and quality assessment of the imaging group (multiple imaging results) of the object.

[0345] Through this identification process, offline configuration, as a prerequisite for online identification, can narrow down the candidate range of formulas. Thus, the online identification process can balance identification efficiency and accuracy. Furthermore, the offline configuration process enables the configuration of sample formulas through simple interaction between the user and the client, thereby reducing the operational threshold, simplifying the configuration process, and improving configuration efficiency and user experience.

[0346] Corresponding to the above image filtering method, embodiments of this application also provide an image filtering device, such as... Figure 12 As shown, the image filtering device may include:

[0347] The first obtaining module 1210 is used to obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0348] The second obtaining module 1220 is used to obtain at least one image feature of the candidate image; wherein the image feature is a feature that can characterize the imaging quality of the content to be identified;

[0349] The result determination module 1230 is used to determine the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the at least one image feature.

[0350] Optionally, the second obtaining module is specifically used for:

[0351] According to the target feature extraction strategy, image features are extracted from the recognition region of the candidate image to obtain at least one image feature of the candidate image; wherein, the target feature extraction strategy is used to characterize the type of image feature to be extracted and / or the feature extraction method used.

[0352] Optionally, the method for determining the target feature extraction strategy includes:

[0353] When target additional information is set for the at least one candidate image, a target feature extraction strategy is generated based on the target additional information; wherein, the target additional information represents: morphological features and / or content information about the content to be identified;

[0354] If no target additional information is set for the at least one candidate image, the default feature extraction strategy will be determined as the target feature extraction strategy.

[0355] Optionally, the at least one image feature includes features extracted from the candidate image based on a preset feature extraction model;

[0356] The preset feature extraction model is a model trained based on image samples and the corresponding label information of the image samples; wherein, the label information is used to characterize the imaging quality of the content to be identified in the image samples.

[0357] Optionally, the device further includes:

[0358] The third obtaining module is configured to obtain reference content of the candidate image before the result determining module determines the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the at least one image feature; and to determine a target auxiliary evaluation index of the candidate image based on the reference content of the candidate image; wherein the reference content is used to evaluate the accuracy of the recognition result of the candidate image; the target auxiliary evaluation index includes a first auxiliary evaluation index and / or a second auxiliary evaluation index; the first auxiliary evaluation index characterizes the degree of consistency between the character content represented by the recognition result of the candidate image and the reference content; and the second auxiliary evaluation index characterizes the confidence level of the reference content of the candidate image.

[0359] The result determination module includes:

[0360] The determining unit is configured to determine an image selection result from the at least one candidate image based on the recognition result of the candidate image, the at least one image feature, and the target auxiliary evaluation index.

[0361] Optionally, the third obtaining module obtains reference content for the candidate image, including:

[0362] If a truth value exists for the candidate image, the truth value is determined as the reference content of the candidate image;

[0363] If the true value does not exist, a reference image of the candidate image is obtained, and image recognition processing is performed on the reference image of the candidate image to obtain the reference content of the candidate image;

[0364] The content to be identified in the reference image of the candidate image is the same as the content to be identified in the candidate image.

[0365] Optionally, the third obtaining module obtains a reference image of the candidate image, including:

[0366] The candidate image is subjected to image transformation processing, and a reference image is determined based on the image obtained after image transformation processing.

[0367] Optionally, the determining unit is specifically used for:

[0368] Based on at least one evaluation metric of the candidate image, a score value for at least one scoring dimension of the candidate image is determined, and based on the score value for at least one scoring dimension of the candidate image, a quality score value for the imaging quality of the content to be identified in the candidate image is analyzed; wherein, the at least one evaluation metric of the candidate image includes: confidence in the recognition result of the candidate image, at least one of the at least one image feature and target auxiliary evaluation metric.

[0369] Based on the quality score of the candidate image, a candidate image is selected from the at least one candidate image to obtain the image filtering result.

[0370] Optionally, the determining unit determines a score value for at least one scoring dimension of the candidate image based on at least one evaluation metric of the candidate image, including:

[0371] According to the scoring strategy corresponding to the evaluation index of the candidate image, the score value of the scoring dimension corresponding to the evaluation index of the candidate image is determined; wherein, different evaluation indexes correspond to different scoring dimensions.

[0372] Optionally, the determining unit determines a score value for at least one scoring dimension of the candidate image based on at least one evaluation metric of the candidate image, including:

[0373] For the candidate image, based on the scoring strategy corresponding to the at least one evaluation index combination, a score value for a scoring dimension corresponding to the evaluation index combination is determined.

[0374] For the candidate image, excluding the combination of evaluation indicators, a score value for a scoring dimension corresponding to the evaluation indicator is determined according to the scoring strategy corresponding to the evaluation indicator.

[0375] Optionally, the scoring strategy corresponding to this combination of evaluation indicators includes:

[0376] Based on the target indicators in this combination of evaluation indicators, an initial score is obtained;

[0377] Based on the other evaluation indicators in this combination of evaluation indicators, excluding the target indicator, determine the score correction coefficient;

[0378] The initial score value is corrected using the score correction coefficient to obtain the score value of a score dimension corresponding to the combination of evaluation indicators.

[0379] This application also provides an image filtering device, such as... Figure 13 As shown, it may include:

[0380] The recognition result acquisition module 1310 is used to obtain the recognition result obtained by performing image recognition processing on at least one candidate image;

[0381] The image content recognition module 1320 is used to perform image recognition processing on the reference image of the candidate image to obtain the reference content of the candidate image; wherein the content to be recognized in the reference image of the candidate image is the same as the content to be recognized in the candidate image.

[0382] The indicator analysis module 1330 is used to determine the target auxiliary evaluation indicator of the candidate image based on the reference content of the candidate image.

[0383] The filtering result determination module 1340 is used to determine the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the target auxiliary evaluation index.

[0384] Optionally, the target auxiliary evaluation index includes a first auxiliary evaluation index and / or a second auxiliary evaluation index; the first auxiliary evaluation index characterizes the degree of consistency between the character content represented by the recognition result of the candidate image and the reference content; the second auxiliary evaluation index characterizes the confidence level of the reference content of the candidate image; and / or,

[0385] The reference image for the candidate image is obtained in the following ways:

[0386] The candidate image is subjected to image transformation processing, and a reference image is determined based on the image obtained after image transformation processing.

[0387] Optionally, the image content recognition module is specifically used to: if there is no ground truth for the candidate image, perform image recognition processing on the reference image of the candidate image to obtain the reference content of the candidate image;

[0388] The device further includes:

[0389] The character content determination module is used to determine the true value as the reference content of the candidate image if the true value exists, and to trigger the index analysis module to determine the target auxiliary evaluation index of the candidate image based on the reference content of the candidate image.

[0390] Optionally, the filtering result determination module is specifically used for:

[0391] Based on at least one evaluation metric of the candidate image, a score value for at least one scoring dimension of the candidate image is determined, and based on the score value for at least one scoring dimension of the candidate image, a quality score value for the imaging quality of the content to be identified in the candidate image is analyzed; wherein, the at least one evaluation metric of the candidate image includes: confidence level and / or target auxiliary evaluation metric in the recognition result of the candidate image;

[0392] Based on the quality score of the candidate image, a candidate image is selected from the at least one candidate image to obtain the image filtering result.

[0393] Based on the above identification method, this application embodiment also provides an identification device applied to a client; such as Figure 14 As shown, the identification device includes:

[0394] The receiving module 1410 is used to receive configuration requests for samples;

[0395] The display module 1420 is configured to respond to the configuration request by displaying a first recognition result and a first evaluation index for the sample; the first evaluation index indicates the imaging quality of the corresponding first recognition result; the first recognition result is a recognition result obtained by performing image recognition processing on multiple imaging results of the sample, and the corresponding first evaluation index satisfies the first screening condition; the multiple imaging results are obtained by imaging the sample based on set multi-group imaging conditions.

[0396] The setting module 1430 is used to respond to user interaction and set the imaging conditions corresponding to the first recognition result to the first imaging conditions for the sample to complete the configuration.

[0397] Optionally, the display module is further configured to display the imaging result and / or imaging information to which the second identification result belongs;

[0398] The imaging information is used to characterize the first imaging condition corresponding to the second recognition result.

[0399] Optionally, the identification device further includes:

[0400] An association module is used to associate the sample with the configuration information in response to the input configuration information before the receiving module receives the configuration request for the sample;

[0401] The configuration information is used for the image recognition processing and / or the generation of the first evaluation index.

[0402] Optionally, the association module, in response to input configuration information, associates the sample with the configuration information, including:

[0403] Display the configuration interface;

[0404] In response to the input of configuration information on the configuration interface, the setting information is associated with the sample.

[0405] Optionally, the association module, in response to input configuration information, associates the sample with the configuration information, including:

[0406] A reference image of the sample is shown;

[0407] In response to an editing operation on the reference image, the edited content corresponding to the editing operation is displayed on the reference image;

[0408] The editing operation includes at least one of the following:

[0409] Draw the identified area; adjust the identified area's position; adjust the identified area's size.

[0410] Optionally, the client provides a truth value editing area for inputting text truth values, and this is a first configuration item that can be omitted.

[0411] Optionally, the client provides a second configuration option for the text location function, which has a user-switchable option to enable or disable it.

[0412] Optionally, the setting module is specifically used for:

[0413] In response to the interactive instruction that confirms the characterization, the imaging conditions corresponding to the first identification result are set to the first imaging conditions for the sample to complete the configuration;

[0414] In response to an interactive instruction to reset the representation, the first identification result for the sample and the first evaluation index corresponding to the first identification result are displayed again.

[0415] In addition, embodiments of this application also provide an electronic device, such as... Figure 15 As shown, it includes:

[0416] Memory 1501 is used to store computer programs;

[0417] The processor 1502 is used to execute the program stored in the memory 1501 to implement the image filtering method and / or recognition method provided in the embodiments of this application.

[0418] Furthermore, the aforementioned electronic device may also include a communication bus and / or a communication interface, with the processor 1502, the communication interface, and the memory 1501 communicating with each other via the communication bus.

[0419] The communication bus mentioned in the aforementioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned electronic device and other devices. The memory can include Random Access Memory (RAM), or Non-Volatile Memory (NVM), such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor. The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0420] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements any of the above-described image filtering and / or recognition methods.

[0421] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the image filtering and / or recognition methods described in the above embodiments.

[0422] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a solid-state drive (SSD), etc.

[0423] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0424] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The above descriptions are merely preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. An image filtering method, characterized in that, include: Obtain the recognition result obtained by performing image recognition processing on at least one candidate image; Obtain at least one image feature of the candidate image; wherein the image feature is a feature capable of characterizing the imaging quality of the content to be identified; Based on the recognition results of the candidate images and the at least one image feature, an image filtering result is determined from the at least one candidate image.

2. The method according to claim 1, characterized in that, The process of obtaining at least one image feature of the candidate image includes: According to the target feature extraction strategy, image features are extracted from the recognition region of the candidate image to obtain at least one image feature of the candidate image; wherein, the target feature extraction strategy is used to characterize the type of image feature to be extracted and / or the feature extraction method used.

3. The method according to claim 2, characterized in that, The method for determining the target feature extraction strategy includes: When target additional information is set for the at least one candidate image, a target feature extraction strategy is generated based on the target additional information; wherein, the target additional information represents: morphological features and / or content information about the content to be identified; If no target additional information is set for the at least one candidate image, the default feature extraction strategy will be determined as the target feature extraction strategy.

4. The method according to claim 1, characterized in that, The at least one image feature includes features extracted from the candidate image based on a preset feature extraction model; The preset feature extraction model is a model trained based on image samples and the corresponding label information of the image samples; wherein, the label information is used to characterize the imaging quality of the content to be identified in the image samples.

5. The method according to any one of claims 1-4, characterized in that, Before determining the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the at least one image feature, the method further includes: Obtain reference content for the candidate image; wherein the reference content is used to evaluate the accuracy of the recognition result of the candidate image; Based on the reference content of the candidate image, a target auxiliary evaluation index is determined for the candidate image; the target auxiliary evaluation index includes a first auxiliary evaluation index and / or a second auxiliary evaluation index; the first auxiliary evaluation index characterizes the degree of consistency between the result content represented by the recognition result of the candidate image and the reference content; the second auxiliary evaluation index characterizes the confidence level of the reference content of the candidate image. The process of determining image filtering results from at least one candidate image based on the recognition result of the candidate image and the at least one image feature includes: Based on the recognition results of the candidate images, the at least one image feature, and the target auxiliary evaluation index, an image screening result is determined from the at least one candidate image.

6. The method according to claim 5, characterized in that, The reference content for obtaining the candidate image includes: If a truth value exists for the candidate image, the truth value is determined as the reference content of the candidate image; If the true value does not exist, a reference image of the candidate image is obtained, and image recognition processing is performed on the reference image of the candidate image to obtain the reference content of the candidate image; The content to be identified in the reference image is the same as the content to be identified in the candidate image.

7. The method according to claim 6, characterized in that, The reference image for obtaining the candidate image includes: The candidate image is subjected to image transformation processing, and a reference image is determined based on the image obtained after image transformation processing.

8. The method according to claim 5, characterized in that, The process of determining image selection results from at least one candidate image based on the recognition result of the candidate image, the at least one image feature, and the target auxiliary evaluation index includes: Based on at least one evaluation metric of the candidate image, a score value for at least one scoring dimension of the candidate image is determined, and based on the score value for at least one scoring dimension of the candidate image, a quality score value for the imaging quality of the content to be identified in the candidate image is analyzed; wherein, the at least one evaluation metric of the candidate image includes: confidence in the recognition result of the candidate image, at least one of the at least one image feature and target auxiliary evaluation metric. Based on the quality score of the candidate image, a candidate image is selected from the at least one candidate image to obtain the image filtering result.

9. The method according to claim 8, characterized in that, Determining the score value of at least one scoring dimension of the candidate image based on at least one evaluation metric includes: According to the scoring strategy corresponding to the evaluation index of the candidate image, the score value of the scoring dimension corresponding to the evaluation index of the candidate image is determined; wherein, different evaluation indexes correspond to different scoring dimensions.

10. The method according to claim 8, characterized in that, Determining the score value of at least one scoring dimension of the candidate image based on at least one evaluation metric includes: For the candidate image, based on the scoring strategy corresponding to the at least one evaluation index combination, a score value for a scoring dimension corresponding to the evaluation index combination is determined. For the candidate image, excluding the combination of evaluation indicators, a score value for a scoring dimension corresponding to the evaluation indicator is determined according to the scoring strategy corresponding to the evaluation indicator.

11. The method according to claim 10, characterized in that, The scoring strategies corresponding to this combination of evaluation indicators include: Based on the target indicators in this combination of evaluation indicators, an initial score is obtained; Based on the other evaluation indicators in this combination of evaluation indicators, excluding the target indicator, determine the score correction coefficient; The initial score value is corrected using the score correction coefficient to obtain the score value of a score dimension corresponding to the combination of evaluation indicators.

12. An image filtering method, characterized in that, include: Obtain the recognition result obtained by performing image recognition processing on at least one candidate image; Image recognition processing is performed on the reference image of the candidate image to obtain the reference content of the candidate image; wherein the content to be identified in the reference image of the candidate image is the same as the content to be identified in the candidate image. Based on the reference content of the candidate image, the target auxiliary evaluation index of the candidate image is determined; Based on the recognition results of the candidate images and the target auxiliary evaluation index, the image screening results are determined from the at least one candidate image.

13. The method according to claim 12, characterized in that, The target auxiliary evaluation index includes a first auxiliary evaluation index and / or a second auxiliary evaluation index; the first auxiliary evaluation index characterizes the degree of consistency between the result content represented by the recognition result of the candidate image and the reference content; The second auxiliary evaluation metric characterizes the confidence level of the reference content of the candidate image; And / or, The methods for obtaining the reference image for the candidate image include: The candidate image is subjected to image transformation processing, and a reference image is determined based on the image obtained after image transformation processing.

14. The method according to claim 12 or 13, characterized in that, The step of performing image recognition processing on the reference image of the candidate image to obtain the reference content of the candidate image includes: If no true value exists for the candidate image, perform image recognition processing on the reference image of the candidate image to obtain the reference content of the candidate image; The method further includes: If the true value exists, the true value is determined as the reference content of the candidate image, and the step of determining the target auxiliary evaluation index of the candidate image based on the reference content of the candidate image is triggered.

15. The method according to claim 12 or 13, characterized in that, The process of determining image selection results from at least one candidate image based on the recognition results of the candidate image and the target-assisted evaluation index includes: Based on at least one evaluation metric of the candidate image, a score value for at least one scoring dimension of the candidate image is determined, and based on the score value for at least one scoring dimension of the candidate image, a quality score value for the imaging quality of the content to be identified in the candidate image is analyzed; wherein, the at least one evaluation metric of the candidate image includes: the confidence level in the recognition result of the candidate image and at least one of the target auxiliary evaluation metrics. Based on the quality score of the candidate image, a candidate image is selected from the at least one candidate image to obtain the image filtering result.

16. A method for identification, characterized in that, Applied to a client, the method includes: Receive configuration requests for the sample; In response to the configuration request, a first recognition result and a first evaluation index for the sample are displayed; the first evaluation index indicates the imaging quality of the corresponding first recognition result; the first recognition result is the recognition result obtained by performing image recognition processing on multiple imaging results of the sample, and the corresponding first evaluation index satisfies the first screening condition; the multiple imaging results are obtained by imaging the sample based on set multi-group imaging conditions. In response to user interaction, the imaging conditions corresponding to the first recognition result are set to the first imaging conditions for the sample to complete the configuration.

17. The method according to claim 16, characterized in that, The method further includes: Receive identification requests for the object to be identified; In response to the recognition request, a second recognition result and a second evaluation index for the object are displayed; the second evaluation index indicates the imaging quality of the corresponding second recognition result; the second recognition result is an recognition result obtained by performing image recognition processing on at least one imaging result of the object, and the corresponding second evaluation index satisfies the second screening condition; the at least one imaging result is obtained by imaging the object based on at least one of the first imaging conditions.

18. The method according to claim 17, characterized in that, The method further includes: Display the imaging result to which the second identification result belongs, and / or, imaging information used to characterize the first imaging condition corresponding to the second identification result.

19. The method according to any one of claims 16-18, characterized in that, Before receiving the configuration request for the sample, the method further includes: In response to input configuration information, associate the configuration information with the sample; The configuration information is used for the image recognition processing and / or the generation of the first evaluation index.

20. The method according to claim 19, characterized in that, In response to input configuration information, the configuration information is associated with the sample, including: Display the configuration interface; In response to the input of configuration information on the configuration interface, the setting information is associated with the sample.

21. The method according to claim 19, characterized in that, The step of responding to input configuration information and associating the sample with the configuration information includes: A reference image of the sample is shown; In response to an editing operation on the reference image, the edited content corresponding to the editing operation is displayed on the reference image; The editing operation includes at least one of the following: Draw the identified area; adjust the identified area's position; adjust the identified area's size.

22. The method according to any one of claims 16-18, characterized in that, The client provides a truth value editing area for inputting truth values, and this area is a first configuration item that can be omitted.

23. The method according to any one of claims 16-18, characterized in that, The client provides a second configuration option for the location function of the content to be identified, which has a user-switchable option to enable or disable it.

24. The method according to any one of claims 16-18, characterized in that, The step of responding to user interaction by setting the imaging conditions corresponding to the first recognition result to the first imaging conditions for the sample to complete the configuration includes: In response to the interactive instruction that confirms the characterization, the imaging conditions corresponding to the first identification result are set to the first imaging conditions for the sample to complete the configuration; In response to an interactive instruction to reset the representation, the first identification result for the sample and the first evaluation index corresponding to the first identification result are displayed again.

25. An image filtering device, characterized in that, include: The first obtaining module is used to obtain the recognition result obtained by performing image recognition processing on at least one candidate image; The second obtaining module is used to obtain at least one image feature of the candidate image; wherein the image feature is a feature that can characterize the imaging quality of the content to be identified; The result determination module is used to determine the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the at least one image feature.

26. An image filtering device, characterized in that, include: The recognition result acquisition module is used to obtain the recognition result obtained by performing image recognition processing on at least one candidate image; An image content recognition module is used to perform image recognition processing on a reference image of the candidate image to obtain the reference content of the candidate image; wherein the content to be recognized in the reference image is the same as the content to be recognized in the candidate image. The indicator analysis module is used to determine the target auxiliary evaluation indicators of the candidate image based on the reference content of the candidate image. The filtering result determination module is used to determine the image filtering result from the at least one candidate image based on the recognition result of the candidate image and the target auxiliary evaluation index.

27. An identification device, characterized in that, Applied to a client, the device includes: The receiving module is used to receive configuration requests for samples; The display module is used to respond to the configuration request and display a first recognition result and a first evaluation index for the sample; the first evaluation index indicates the imaging quality of the corresponding first recognition result; the first recognition result is the recognition result obtained by image recognition processing of multiple imaging results of the sample, and the corresponding first evaluation index satisfies the first screening condition; the multiple imaging results are obtained by imaging the sample based on set multi-group imaging conditions. The setting module is used to respond to user interaction and set the imaging conditions corresponding to the first recognition result to the first imaging conditions for the sample to complete the configuration.

28. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method described in any one of claims 1-24.

29. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-24.

Images