Image processing method, device and system

Abstract

The present application discloses an image processing method, device and system. The method comprises: receiving an image set, wherein the image set comprises: multiple images and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; determining a target image in the image set based on the weight values ​​of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted; and outputting the target image. The present application solves the problem in the related art that the training samples may contain label errors and low-quality samples, thereby increasing the difficulty of using training samples to train the face recognition network.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to an image processing method, device and system. Background Art

[0002] In the era of big data, image recognition networks are trained using various types of training samples to obtain image recognition models. However, current training samples used to train image recognition networks often have some issues. For example, related techniques often use training samples with incorrect labels and low-quality samples, resulting in low recognition accuracy for image recognition models trained using these training samples.

[0003] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0004] The embodiments of the present application provide an image processing method, device and system to at least solve the technical problem in the related art that the image recognition model trained using the training samples has low recognition accuracy due to the presence of label errors and low-quality samples in the training samples.

[0005] According to one aspect of an embodiment of the present application, an image processing method is provided, comprising: receiving an image set, wherein the image set comprises: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; determining a target image in the image set based on the weight values ​​of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted; and outputting the target image.

[0006] According to another aspect of an embodiment of the present application, an image processing method is also provided, including: displaying an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to characterize the importance of the first label; displaying a target image in the image set, wherein the target image is determined based on the weight values ​​of the multiple images, and the first label corresponding to the target image is the label to be adjusted.

[0007] According to another aspect of an embodiment of the present application, an image processing method is also provided, including: receiving an image by calling a first interface, wherein the first interface includes: a first parameter and a second parameter, the parameter value of the first parameter is the image, and the parameter value of the second parameter is a first label corresponding to the image; determining a weight value of the image, wherein the weight value is used to characterize the importance of the first label; determining a recognition result of the image based on the weight value of the image, wherein the recognition result is used to characterize whether the image is a target image, and the first label corresponding to the target image is a label to be adjusted; outputting the recognition result by calling a second interface, wherein the second interface includes: a third parameter, and the parameter value of the third parameter is the recognition result.

[0008] According to another aspect of an embodiment of the present application, an image processing method is also provided, including: receiving a product image set, wherein the product image set includes: multiple product images, and a first product label corresponding to each product image; determining weight values ​​of the multiple product images, wherein the weight values ​​are used to characterize the importance of the first product label; based on the weight values ​​of the multiple product images, determining a target product image in the product image set, wherein the first product label corresponding to the target product image is the label to be adjusted; and outputting the target product image.

[0009] According to another aspect of an embodiment of the present application, an image processing device is also provided, including: a receiving module for receiving an image set, wherein the image set includes: multiple images and a first label corresponding to each image; a first determination module for determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; a second determination module for determining a target image in the image set based on the weight values ​​of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted; and an output module for outputting the target image.

[0010] According to another aspect of an embodiment of the present application, an image processing device is also provided, including: a first display module for displaying an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; a determination module for determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; a second display module for displaying a target image in the image set, wherein the target image is determined based on the weight values ​​of the multiple images, and the first label corresponding to the target image is the label to be adjusted.

[0011] According to another aspect of an embodiment of the present application, an image processing device is also provided, including: a first calling module, used to receive a collection by calling a first interface, wherein the first interface includes: a first parameter and a second parameter, the parameter value of the first parameter is the image, and the parameter value of the second parameter is the first label corresponding to the image; a first determination module, used to determine the weight value of the image, wherein the weight value is used to characterize the importance of the first label; a second determination module, used to determine the recognition result of the image based on the weight value of the image, wherein the recognition result is used to characterize whether the image is a target image, and the first label corresponding to the target image is the label to be adjusted; a second calling module, used to output the recognition result by calling the second interface, wherein the second interface includes: a third parameter, and the parameter value of the third parameter is the recognition result.

[0012] According to another aspect of an embodiment of the present application, an image processing device is also provided, including: a receiving module for receiving a product image set, wherein the product image set includes: multiple product images, and a first product label corresponding to each product image; a first determination module for determining weight values ​​of the multiple product images, wherein the weight value is used to characterize the importance of the first product label; a second determination module for determining a target product image in the product image set based on the weight values ​​of the multiple product images, wherein the first product label corresponding to the target product image is the label to be adjusted; and an output module for outputting the target product image.

[0013] According to another aspect of an embodiment of the present application, a computer-readable storage medium is further provided. The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned image processing method.

[0014] According to another aspect of an embodiment of the present application, a computer terminal is provided, including: a memory and a processor, wherein the processor is configured to run a program stored in the memory, wherein the above-mentioned image processing method is executed when the program is run.

[0015] According to another aspect of an embodiment of the present application, an image processing system is also provided, including: a processor; and a memory connected to the processor, for providing the processor with instructions for processing the following processing steps: receiving an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to characterize the importance of the first label; determining a target image in the image set based on the weight values ​​of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted; and outputting the target image.

[0016] In an embodiment of the present application, after receiving a plurality of images and a first label corresponding to each image, weight values ​​for representing the importance of the first label of the plurality of images can be determined, and then based on the weight values ​​of the plurality of images, a target image in the image set can be determined, wherein the first label corresponding to the target image is the label to be adjusted, and finally the target image is output, thereby achieving the purpose of accurately locating the target image in the image set. It is easy to note that after receiving the image set, a target image with greater uncertainty can be determined based on the weight values ​​of the plurality of images, that is, an image with lower quality and incorrect label can be determined in the plurality of images. Since the label corresponding to the target image with greater uncertainty is more likely to be incorrect, the error rate of the label in the image set can be reduced by adjusting the first label corresponding to the target image, so that the image recognition network can be trained using the adjusted image set to obtain an image recognition model with higher recognition accuracy, thereby achieving the technical effect of improving the recognition accuracy of the image recognition model, thereby solving the technical problem in the related art that the image recognition model trained using the training samples has low recognition accuracy due to the presence of incorrect labels and low-quality samples in the training samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 This is a hardware structure block diagram of a computer terminal (or mobile device) for implementing an image processing method according to an embodiment of the present application;

[0019] Figure 2 is a flowchart of an image processing method according to Example 1 of the present application;

[0020] Figure 3 is a schematic diagram of an interactive interface according to Example 1 of the present application;

[0021] Figure 4 is a flowchart of an optional image processing method according to Example 1 of the present application;

[0022] Figure 5 is a flowchart of another optional image processing method according to Example 1 of the present application;

[0023] Figure 6 is a flowchart of another optional image processing method according to Example 1 of the present application;

[0024] Figure 7ais a flowchart of another optional image processing method according to Example 1 of the present application;

[0025] Figure 7b This is a schematic diagram of experimental data for error label repair according to Example 1 of the present application;

[0026] Figure 8 is a flowchart of an image processing method according to Example 2 of the present application;

[0027] Figure 9 is a schematic diagram of an image processing device according to Embodiment 3 of the present application;

[0028] Figure 10 is a schematic diagram of an image processing device according to Embodiment 4 of the present application;

[0029] Figure 11 is a flowchart of an image processing method according to Example 5 of the present application;

[0030] Figure 12 is a flowchart of an image processing method according to Example 6 of the present application;

[0031] Figure 13 is a schematic diagram of an image processing device according to Embodiment 7 of the present application;

[0032] Figure 14 is a schematic diagram of an image processing device according to Example 8 of the present application;

[0033] Figure 15 This is a structural block diagram of a computer terminal according to embodiment 10 of the present application. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0035] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following interpretations:

[0037] Uncertain sample learning: Uncertain samples include uncertainty in labels and uncertainty caused by low sample quality.

[0038] Self-healing network: A mechanism that determines whether to change labels based on the distribution of logits (unnormalized probabilities), where the unnormalized probabilities are the weighted sum of the features.

[0039] The current solution is mainly based on sample clustering and feature fusion methods to reduce label errors and low-quality samples; specifically, the sample clustering-based method first clusters the samples, and then divides the samples into three levels: high, medium and low according to the sample clustering density, and then adds them to the network for training respectively. This approach requires clustering in advance, which has high computational overhead, and the accuracy of the clustered samples cannot be determined; the feature fusion-based method first uses clean samples to fuse features with uncertain samples, and then aggregates them into one feature, hoping that the clean samples will have a larger weight to offset the impact of uncertain samples. However, this approach will only destroy the diversity of data set samples, and can only work in scenarios with a low proportion of uncertain samples, and its limitations are relatively large.

[0040] In order to solve the above problems, this application provides the following implementation solution, which can automatically identify uncertain samples and repair them. It can reduce label errors and low-quality samples without affecting sample diversity, and at the same time, it can also reduce the cost of processing samples.

[0041] Example 1

[0042] According to an embodiment of the present application, an embodiment of a method for image processing is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0043] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing an image processing method. Figure 1 As shown, the computer terminal 10 (or mobile device 10) may include one or more (illustrated as 102a, 102b, ..., 102n) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0044] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be fully or partially integrated into any of the other components of the computer terminal 10 (or mobile device). The data processing circuitry may function as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0045] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the image processing method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned image processing method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0046] The transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0047] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0048] It should be noted that, in some optional embodiments, the above Figure 1 The computer device (or mobile device) shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the aforementioned computer device (or mobile device).

[0049] Under the above operating environment, this application provides Figure 2 The image processing method shown. Figure 2 This is a flow chart of an image processing method according to Example 1 of the present application. Figure 2 As shown, the method may include the following steps:

[0050] Step S202: receiving an image set.

[0051] The image set includes: multiple images and a first label corresponding to each image.

[0052] It should be noted that there may be low-quality images and images with inaccurate labels in the image collection. If the image collection is directly used as training samples to train the image recognition model, the recognition accuracy of the image recognition model will be reduced and it will not be able to meet actual recognition needs.

[0053] The application scenarios for receiving the image collection in the above steps can be e-commerce scenarios, medical scenarios, educational scenarios, conference scenarios, as well as scenarios such as banking and security that require facial recognition, but are not limited to these. In e-commerce scenarios, the above images can be the facial images of a host in a live video broadcast, and the image collection can be a collection of facial images of multiple hosts; the above images can also be product images, and the image collection can be a collection of product images of multiple products. In medical scenarios, the above images can be facial images of medical personnel such as doctors and nurses, and the image collection can be a collection of facial images of multiple medical personnel; the above images can also be medical images, and the image collection can be a collection of medical images of multiple patients; the above images can also be medical record images, and the image collection can be a collection of medical record images of multiple patients. In educational scenarios, the above images can be the facial images of teachers in live education broadcasts, and the image collection can be a collection of facial images of multiple teachers; the above images can also be test paper images, and the image collection can be a collection of test paper images of multiple students; the above images can also be images of teachers' writing on the blackboard, and the image collection can be a collection of images of multiple teachers' writing on the blackboard. In a conference scenario, the aforementioned images could be the facial images of each participant in a video conference, and the image collection could be facial images of different people within a company. In scenarios requiring facial recognition, such as banking and security, the aforementioned images could be facial images of people requiring facial recognition, and the image collection could be a collection of facial images captured by cameras in banks and security areas.

[0054] The first label in the above steps can be the labeling result obtained by the labeler manually labeling each image. The specific type of label is different for different image types. For example, for a product image of a commodity, the first label can indicate the commodity type of the commodity; for a medical image, the first label can indicate the examination type, such as CT examination, B-ultrasound examination, etc., or the type of diagnosis result, such as normal, presence of cyst, etc.; for an examination paper image, the first label can be the subject type of the examination paper, such as Chinese, mathematics, English, etc., or the type of question on the examination paper, such as objective questions, subjective questions, etc.; for a blackboard image, the first label can be the subject type corresponding to the blackboard, such as Chinese, mathematics, English, etc., or the specific type of blackboard content, such as knowledge points, examples, etc.; for a face image, the first label can indicate different facial movements in the face image, such as opening the mouth, opening the eyes, closing the eyes, etc., or different expressions, such as happy, sad, fearful, etc.

[0055] Multiple tags can be pre-set for different images, allowing the annotator to select the appropriate tag during the annotation process. For example, taking a product image as an example, seven tags can be pre-set to represent the seven categories: apparel, beauty, department stores, electrical appliances, luggage, fresh food, and medical. The annotator can select the corresponding tag based on the product image. For example, if the product is a dress, the annotator can use the tag corresponding to clothing.

[0056] It should be noted that to reduce the amount of data in the label content, corresponding numbers can be set in advance for different label contents. For example, still using the product type of the product image as an example, the seven pre-set labels can be labeled with the numbers 1, 2, 3, 4, 5, 6, and 7 respectively. That is, label 1 represents clothing, label 2 represents beauty products, label 3 represents department stores, label 4 represents electrical appliances, label 5 represents luggage, label 6 represents fresh food, and label 7 represents medical care. If the product in a product image is a dress, the first label labeled for the product image is label 1.

[0057] The first tag in the above step may be recorded in the attribute information of the image, or may be marked on the image.

[0058] In an optional embodiment, in order to better process the image set, the acquired image set can be transmitted to a corresponding processing device for processing, for example, directly transmitted to a user's computer terminal (e.g., a laptop computer, personal computer, etc.) for processing, or transmitted to a cloud server via the user's computer terminal for processing. It should be noted that because the processing of images in the image set requires a large amount of computing resources, the embodiments of this application are described using a cloud server as an example processing device.

[0059] For example, in order to facilitate users to upload image collections, an interactive interface can be provided to users, such as Figure 3 As shown, users can click the "Select Image" button to select images to be processed from a large number of stored images, or select multiple images in batches, generate an image set based on the selected images, and upload the image set to the cloud server for processing by clicking the "Upload" button. In addition, to facilitate users to confirm whether the images in the generated image set are the images that need to be processed, the images in the generated image set can be displayed in the "Image Display" area. After the user confirms that they are correct, they can click the "Upload" button to upload the data.

[0060] It should be noted that data can be exchanged between the client and the cloud server through a specific interface. The client can pass each image in the image set selected by the user and the first tag corresponding to the image into the interface function as two parameters of the interface, thereby achieving the purpose of uploading the image set to the cloud server.

[0061] Step S204: Determine weight values ​​of the multiple images.

[0062] The weight value is used to represent the importance of the first tag.

[0063] In the above steps, the smaller the image weight, the lower the image quality and the lower the accuracy of the corresponding first label. For example, a blurry image or an obstruction indicates that the image quality is low, resulting in a lower image weight. Lower-quality images have lower accuracy of the first label, less importance of the first label, and thus a lower image weight.

[0064] In an optional embodiment, a neural network model may be pre-trained to extract features of each image, and a weight value corresponding to each image may be determined based on the similarity between the extracted features and the reference features.

[0065] Specifically, a convolutional neural network can be used as the backbone network, and the convolutional neural network can be used to extract N image features of each image, where the dimension of the image features can be D. Then, a fully connected layer is used to classify the N image features, and finally an activation function is used to predict the image features to obtain the weight value of each image; wherein, the fully connected layer plays the role of a classifier in the entire neural network model; the activation function can be a sigmoid function (S-shaped growth curve function).

[0066] In another optional embodiment, multiple pre-trained neural networks can be used to determine the weight of each image. This allows the weight of the image to be determined from multiple perspectives, preventing significant deviations in the results obtained by a single neural network. For example, two pre-trained neural networks can be used to determine the weight of each image.

[0067] Step S206 : determining a target image in the image set based on the weight values ​​of the multiple images.

[0068] The first label corresponding to the target image is the label to be adjusted.

[0069] In an optional embodiment, after obtaining the weight values ​​of multiple images, the images can be sorted from large to small according to their corresponding weight values, and one or more images with lower weight values ​​can be obtained as target images. Since the weight value corresponding to the target image is relatively low, it can be determined that the accuracy of the first label corresponding to the target image is relatively low. At this time, the first label corresponding to the target image needs to be adjusted to improve the accuracy of the first label corresponding to the image.

[0070] In another optional embodiment, when multiple neural network models determine weight values ​​for multiple images, the weight values ​​corresponding to the multiple images in each neural network model can be sorted from large to small, and at least one image with a higher weight value ranking in the multiple neural network models can be exchanged to increase the diversity of samples in each neural network model, and the multiple neural network models can be trained based on the at least one image with a higher weight value ranking to improve the accuracy of the multiple neural network models.

[0071] Step S208: output the target image.

[0072] In an optional embodiment, the output target image can be identified, the probability values ​​of multiple labels corresponding to the target image can be determined, and the first label can be changed to the label with the highest probability value among the multiple labels. For example, still taking the product image as an example, after the target image is identified, the probability values ​​of the seven pre-set labels can be determined, that is, the probability values ​​of the seven pre-set types can be determined. The higher the probability value, the more the type conforms to the product in the target image. Therefore, the label corresponding to the maximum probability value can be directly used as the actual label of the product in the target image, that is, the first label of the target image can be changed to the label corresponding to the maximum probability value. If the first label annotated by the labeler for the product image is clothing, and the label corresponding to the maximum probability value is luggage, then the first label corresponding to the product image can be changed from clothing to luggage.

[0073] Due to the image itself and recognition accuracy, the probability values ​​of multiple labels are often close. For example, a product image may contain both the main product and ancillary items. Therefore, the same product image may resemble both clothing and luggage. In this case, the probability values ​​of the clothing and luggage labels are relatively close. If the first label is clothing, and the label with the highest probability value is luggage, changing the first label will not significantly improve the target image and will waste computing resources.

[0074] In another optional embodiment, in order to avoid the above situation, the output target image can be identified, and the probability values ​​of multiple labels corresponding to the target image can be determined, wherein there is a first label among the multiple labels, the maximum probability value among the multiple labels is determined, and the difference between the probability value and the probability value of the first label is obtained. If the difference is greater than a preset threshold (the preset threshold can be set according to actual needs, and this application does not make specific restrictions on this), it means that the label manually marked by the labeler is significantly different from the actual label, that is, the first label is an incorrect label. Therefore, the first label can be changed to a label with the largest probability value to reduce the uncertainty of the image; if the difference is less than or equal to the preset threshold, it means that the label manually marked by the labeler is relatively small compared to the actual label, that is, the accuracy of the first label is relatively high, and at this time, there is no need to change the first label corresponding to the target image. For example, still taking the product image as an example, after the target image is identified, the probability values ​​of the seven pre-set labels can be determined. If the first label annotated by the labeler for the product image is clothing, and the label corresponding to the maximum probability value is luggage, the difference in the probability values ​​of the two labels can be calculated. If the difference is greater than the preset threshold, the first label corresponding to the face image can be changed from clothing to luggage; if the difference is less than or equal to the preset value, the first label can be left unchanged, and at this time, the first label is still clothing.

[0075] In an optional embodiment, after the cloud server outputs the target image to the client, the client can output the target image to a user-operable display screen, allowing the user to actively change the first label of the target image. The client can also transmit the target image and the changed label to the user-operable display screen, allowing the user to verify whether the changed label is accurate. If it is accurate, the user can press a confirmation button on the display screen; if it is inaccurate, the user can press a change button on the display screen. At this time, the user can select the label with the highest degree of match with the target image from the multiple provided labels as the changed label. In addition, the user can also add a new label as the changed label and send the changed label to the cloud server.

[0076] For example, in an e-commerce scenario, the first label may be the product information of a product, wherein the product information may include the name of the product, the type of the product, etc. If a product image with inaccurate product information is used to train the product recognition model, the user will obtain inaccurate product information after using the recognition model to identify the product, thereby causing the product search to fail. Therefore, before using a product image set to train the product recognition model, it is necessary to ensure the accuracy of the first label corresponding to each product image in the product image set; based on the weight values ​​corresponding to multiple product images in the product image set, a target product image with lower quality can be determined. The accuracy of the first label corresponding to a target product image with lower quality is generally lower. Therefore, the first label in the target product image can be changed to reduce the uncertainty of the product image in the product image set. Using samples with lower uncertainty to train the product recognition model can improve the accuracy of the product recognition model, so that users can obtain accurate product information when using the product recognition model to identify the product, thereby improving the success rate of the product search.

[0077] For example, in a medical scenario, the first label can be the medical information of a medical image, where the medical information can be the examination type, diagnosis result, etc. If a medical image with inaccurate medical information is used to train the image recognition model, the image recognition model will incorrectly label the medical information of the medical image, causing trouble for doctors and patients. Therefore, before using a medical image set to train the image recognition model, it is necessary to ensure the accuracy of the first label corresponding to each medical image in the medical image set; based on the weight values ​​of multiple medical images in the medical image set, a target medical image with lower quality can be determined. The accuracy of the first label corresponding to the target medical image with lower quality is generally lower. Therefore, the first label in the target medical image can be changed to reduce the uncertainty of the medical image in the medical image set. Using samples with lower uncertainty to train the image recognition model can improve the accuracy of the image recognition model, so that the medical information of the medical image can be quickly identified in the medical scenario and marked in the area corresponding to the lesion, so that the doctor can quickly confirm the patient's medical information.

[0078] Through the solution provided by the above-mentioned embodiment of the present application, after receiving a plurality of images and a first label corresponding to each image, weight values ​​for representing the importance of the first label of the plurality of images can be determined, and then based on the weight values ​​of the plurality of images, a target image in the image set can be determined, wherein the first label corresponding to the target image is the label to be adjusted, and finally the target image is output, thereby achieving the purpose of accurately locating the target image in the image set. It is easy to note that after receiving the image set, the target image with a large uncertainty in the first label can be determined based on the weight values ​​of the plurality of images, that is, the image with low quality and incorrect label can be determined in the plurality of images. Since the label corresponding to the target image with large uncertainty is more likely to be incorrect, the error rate of the label in the image set can be reduced by adjusting the first label corresponding to the target image. The image recognition network can then be trained using the adjusted image set to obtain an image recognition model with high recognition accuracy, thereby achieving the technical effect of improving the recognition accuracy of the image recognition model, thereby solving the technical problem in the related art that the recognition accuracy of the image recognition model obtained by training using the training samples is low due to the presence of incorrect labels and low-quality samples in the training samples.

[0079] In the above embodiment of the present application, after determining the weight values ​​of multiple images, the method also includes: identifying multiple images to obtain recognition results corresponding to the multiple images; obtaining a loss value for each image based on the recognition result and the first label corresponding to each image; and determining the target image based on the weight value and / or loss value of each image.

[0080] The loss value in the above steps can be obtained through the loss function in the neural network, wherein the smaller the loss value, the higher the importance corresponding to the first label, that is, the higher the matching degree between the first label and the image.

[0081] It should be noted that the loss function is used to evaluate the quality of the model, that is, the degree to which the predicted value is different from the true value. For example, to recognize an image, there will be a gap between the recognized result and the result of the first label. This gap can be measured by the loss function. The smaller the value of the loss function, the higher the accuracy of the first label; the larger the value of the loss function, the lower the accuracy of the first label. At this time, the first label needs to be adjusted.

[0082] In addition to directly determining the target image based on the weight value of each image, in an optional embodiment, a pre-trained convolutional neural network can be used to recognize multiple images, obtain recognition results corresponding to the multiple images, and use a loss function to calculate the loss value of each image, and determine the target image based on the loss value of each image, and the loss values ​​can be sorted from high to low, and a preset percentage of images ranked at the top can be used as target images, and the first label corresponding to the target image can be adjusted. In another optional embodiment, a pre-trained convolutional neural network can be used to recognize multiple images, obtain recognition results corresponding to the multiple images, and use a loss function to calculate the loss value of each image, and calculate the score of each image based on the weight value and loss value of each image, and determine the target image based on the score of each image, and the scores can be sorted from high to low, and a preset percentage of images ranked at the bottom can be used as target images, and the first label corresponding to the target image can be adjusted.

[0083] In the above embodiment of the present application, determining the target image based on the weight value and loss value of each image includes: obtaining a score for each image based on the weight value and loss value of each image; and determining the target image based on the scores of multiple images.

[0084] In an optional embodiment, after calculating the loss value of each image, it can be determined that the higher the weight value, the higher the accuracy of the first label, and the larger the loss value, the lower the accuracy of the first label. Therefore, the loss value can be normalized first, and the formula: weight value + (1-normalized loss value) can be used to obtain the score of each image. The higher the score obtained, the higher the quality of the image; the lower the score obtained, the lower the quality of the image. At this time, the scores can be sorted from high to low, and the images with a preset percentage of rankings at the bottom can be used as target images, and the first label corresponding to the target image can be adjusted.

[0085] In another optional embodiment, the loss value can be normalized first, and the formula: (1-weight value) + normalized loss value can be determined to obtain a score for each image. The lower the score, the higher the image quality; the higher the score, the lower the image quality. At this time, the scores can be sorted from high to low, and a preset percentage of images ranked at the top can be used as target images, and the first label corresponding to the target image can be adjusted.

[0086] It should be noted that the above-mentioned preset percentage can be set according to actual needs, for example, it can be 30%, but is not limited thereto.

[0087] In the above embodiment of the present application, based on the weight value and loss value of each image, obtaining the score of each image includes: normalizing the loss value of each image to obtain the normalized result of each image; obtaining the difference between the preset value and the weight value of each image to obtain the weight result of each image; obtaining the sum of the normalized result and the weight result of each image to obtain the score of each image.

[0088] The normalization in the above steps is used to limit the loss value of each image to a certain range, thereby eliminating the adverse effects caused by singular sample data. In the embodiment of the present application, the normalization result can be between 0 and 1, where 1 represents an image with a higher loss value and 0 represents an image with a lower loss value.

[0089] The preset value in the above steps may be 1, and the difference between 1 and the weight value may be obtained to obtain the weight result of each image as (1-weight value).

[0090] Furthermore, the sum of the normalized loss value and (1-weight value) is calculated to obtain the score of each image, and then the scores of each image are sorted from large to small, and the images with a preset percentage of the highest scores are used as target images.

[0091] In the above embodiment of the present application, determining the target image based on the scores of multiple images includes: sorting the multiple images in descending order of their scores; and obtaining a preset number of images with the highest scores to obtain the target image.

[0092] The preset number in the above steps can be set by the user, and can also be set according to the scores of multiple images. For example, when the overall scores of multiple images are high, it means that the overall quality of the multiple images is low. At this time, the preset number can be increased to reduce the uncertainty of the images in the image set. When the overall scores of multiple images are low, it means that the overall quality of the multiple images is high. At this time, the preset number can be reduced to reduce the steps of changing the first label, thereby improving the efficiency of image processing.

[0093] In an optional embodiment, after obtaining the score of each image based on the sum of the normalization result and the weight result of each image, the multiple images can be sorted from large to small according to their scores. The larger the score, the lower the quality of the image and the lower the accuracy of the corresponding first label. At this time, a preset number of images with the highest ranking can be obtained to obtain the target image.

[0094] In the above embodiment of the present application, determining the weight values ​​of multiple images includes: extracting features from the multiple images to obtain feature information corresponding to the multiple images; and processing the feature information of the multiple images to obtain weight values ​​of the multiple images.

[0095] The feature information in the above steps may be attributes of a specific object in the image. For example, taking the product in the product image as an example, the feature information may be the color, size, material, shape, etc. of the product, but is not limited thereto.

[0096] In an optional embodiment, a feature extraction network can be used to extract features from multiple images to obtain feature information corresponding to the multiple images; the feature information is then subjected to redundancy processing and noise processing to make the obtained feature information more accurate, and then the feature information is classified using a fully connected layer, and the sigmoid function is used to determine the weight value of the image corresponding to the classified feature information.

[0097] In the above embodiment of the present application, determining the target image in the image set based on the weight values ​​of multiple images includes: sorting the multiple images in descending order of the weight values ​​of the multiple images; obtaining a preset number of images at the bottom of the sorting to obtain the target image.

[0098] In an optional embodiment, multiple images can be sorted from large to small according to their weight values. At this time, the weight value of the image in the lower order is lower, that is, the quality of the image is lower and its uncertainty is relatively large. Therefore, the accuracy of the first label is low. At this time, the first label can be adjusted to improve the accuracy of the first label, thereby improving the accuracy of the image set.

[0099] In the above embodiment of the present application, the feature information of the multiple images is processed using constraint conditions to obtain weight values ​​of the multiple images.

[0100] The constraints in the above steps can be L RR =max{0,δ1-(α H -α L )}; where L RR Used to express the constraint condition, δ1 is a preset value, α H It is used to represent the mean of the image group with higher weight value, α L The mean of the group of images representing the lower weighted values.

[0101] By processing multiple images with constraint conditions, the similarities of the multiple images can be made different, thereby obtaining multiple weight values ​​corresponding to the multiple images.

[0102] In the above embodiment of the present application, after determining the target image in the image set based on the weight values ​​of multiple images, the method also includes: obtaining a first recognition result of the target image, wherein the first recognition result includes: multiple labels corresponding to the target image, and a probability value corresponding to each label, the multiple labels including the first label; determining the second label corresponding to the maximum probability value in the recognition result; and changing the first label corresponding to the target image to the second label.

[0103] The multiple labels in the above steps may be different labels pre-set for different types of images.

[0104] The first recognition result in the above step can be obtained by using a pre-trained recognition model to identify the target image. The higher the probability value corresponding to the label, the higher the match between the label and the target image; the lower the probability value corresponding to the label, the lower the match between the label and the target image. Therefore, after identifying the probability values ​​corresponding to multiple labels, the label corresponding to the maximum probability value can be directly used as the actual label of the target image. In other words, the first label corresponding to the target image is changed to the label corresponding to the maximum probability value.

[0105] For example, for a product image, assuming that the first label is clothing, the target image can be identified, and the first identification result obtained may include: clothing, beauty, department store, electrical appliances, luggage, fresh food, medical and other multiple labels, as well as probability values ​​corresponding to clothing, beauty, department store, electrical appliances, luggage, fresh food, and medical. Furthermore, it can be determined that the second label corresponding to the maximum probability value in the first identification result is luggage, indicating that the actual type of the product in the target image is luggage. At this time, the first label corresponding to the target image can be changed to luggage to ensure the accuracy of the first label.

[0106] In an optional embodiment, after the first label is changed to the second label, the second label may be transmitted to a user-operable display screen for confirmation by the user, thereby ensuring a high accuracy of the changed second label.

[0107] In another optional embodiment, after obtaining the first recognition result of the target image, the recognition result can be transmitted to a user-operable display screen, so that the user can change the first label according to the probability value corresponding to each label. At this time, the user can change the first label to the second label. The user can also change the first label to other labels that the user considers to be more accurate, for example, it can be a newly added label, so as to increase the flexibility of changing the first label, thereby making the accuracy of the first label higher.

[0108] In the above embodiment of the present application, before changing the first label corresponding to the target image to the second label, the method also includes: obtaining the difference between the probability value corresponding to the second label and the probability value corresponding to the first label; judging whether the difference is greater than a preset value; if the difference is greater than the preset value, changing the first label corresponding to the target image to the second label.

[0109] The preset values ​​in the above steps can be set by the user, and this application does not impose any specific restrictions on this.

[0110] In an optional embodiment, the difference between the probability value corresponding to the second label and the probability value corresponding to the first label can be obtained. If the difference is greater than a preset value, it means that the accuracy of the first label is low. At this time, the first label needs to be changed, and the first label can be changed to the second label; if the difference is less than or equal to the preset value, it means that the accuracy of the first label is high. At this time, if the accuracy requirement is not high, the first label can be left unchanged to reduce the loss of computing resources. If the accuracy requirement is relatively high, the first label can be changed to improve the accuracy of the first label.

[0111] In the above embodiment of the present application, the recognition model is trained using an image set, wherein the loss value of the recognition model is determined based on the loss values ​​of other images in the image set except the target image.

[0112] The recognition model in the above steps can be an image recognition model, which is mainly constructed through a neural network, such as ResNet (residual network), VGG (convolutional network), etc.

[0113] In an optional embodiment, the loss value of the recognition model is determined by the loss values ​​of other images in the image set except the target image, so that the loss function in the recognition model can be optimized, making the recognition result of the recognition model more accurate.

[0114] In the above embodiment of the present application, when a confirmation instruction is received, a first recognition result of the target image is obtained, wherein the confirmation instruction is used to determine to adjust the first label.

[0115] In an optional embodiment, the cloud server can send the determined target image to the terminal device used by the user. After the user clicks confirm on the display screen of the terminal device, the terminal device can generate a confirmation instruction and send the confirmation instruction to the cloud server. After receiving the confirmation instruction, the cloud server obtains the first recognition result of the target image and adjusts the first label corresponding to the target image.

[0116] In another optional implementation, the user can change the target image sent by the cloud server. If the user needs to change the target image, the user can click the change button on the display screen to change the target image. At this time, the terminal device can generate a change instruction and send the change instruction and the changed target image to the cloud server. After receiving the change instruction and the changed target image, the cloud server can obtain the recognition result of the changed target image according to the change instruction.

[0117] In the above embodiment of the present application, after outputting the target image, the method further includes: receiving a third tag corresponding to the target image when a confirmation instruction is received; and changing the first tag corresponding to the target image to the third tag.

[0118] In an optional embodiment, the cloud server can send the target image to the terminal device used by the user. At this time, if the user believes that the first label of the target image is indeed wrong, the user can label the target image with a new label, that is, a third label. At this time, the terminal device can send the third label to the cloud server, and the cloud server can change the first label corresponding to the target image to the third label, so that the label of the target image is more accurate, thereby improving the overall quality of the image collection.

[0119] In the above embodiment of the present application, after outputting the target image, the method also includes: receiving an image selected from the target image; obtaining a second recognition result of the selected image, wherein the second recognition result includes: multiple labels corresponding to the selected image, and a probability value corresponding to each label, the multiple labels including the first label; determining the fourth label corresponding to the maximum probability value in the second recognition result; and changing the first label corresponding to the selected image to the fourth label.

[0120] In an optional embodiment, after the cloud server outputs the target image to the client, the client can output the target image to a user-operable display screen, so that the user can filter the target image, filter out images with errors in the first label, and feedback to the cloud server, so that the cloud server can determine the image fed back by the user, that is, the image selected from the target image is an image with low quality and incorrect label, and the cloud server can identify the selected image to achieve the purpose of automatically changing the first label corresponding to the selected image.

[0121] It should be noted that the implementation method of changing the label of the selected image is the same as the implementation method of changing the label of the target image, and will not be described in detail here.

[0122] In the above embodiment of the present application, outputting the target image includes: determining an output mode of the target image based on a weight value of the target image; and outputting the target image according to the output mode of the target object.

[0123] The output mode in the above step may include one of the following: the arrangement order of the target images, and the target labels of the target images.

[0124] In order to allow the user to focus on the target images whose first labels need to be changed, in an optional embodiment, the target images can be sorted based on their weight values, so that the user can determine the images whose first labels need to be changed based on the sorting results, or manually modify the first labels corresponding to the target images based on the sorting results. In another optional embodiment, the target images can be annotated with different tags based on their weight values, for example, using different colors, different fonts, or different transparency methods to annotate the target images.

[0125] The following combination Figure 4 FIG7 takes the application scenario of face image as an example to explain a preferred embodiment of the present application in detail. The method can be executed by a mobile terminal or a server. In the embodiment of the present application, the method is explained by taking the server as an example. Figure 4 As shown, the method may include the following steps:

[0126] Step S401, obtaining a face image set;

[0127] Step S402, using a convolutional neural network and a processing model to determine a weight value for each face image;

[0128] like Figure 5 As shown in the figure, after the image features of each face image are extracted using a convolutional neural network, the image features of each face image can be classified using a processing model composed of a fully connected layer and an S-shaped growth curve function, and the weight value of each face image can be predicted based on the classified image features.

[0129] Specifically, a convolutional neural network can be used to extract the image feature set F of the face image, where F = [X0, X1, ...X N ]∈R D×N , there are N features in the image feature set F, where D is the dimension of the image feature, and the image features are classified using the fully connected layer, and then the sigmoid function is used to obtain the weight of the face image corresponding to the image feature. Specifically, it can be used in, is the weight value corresponding to each image feature in the face image, is the weight value corresponding to the face image, and σ is used to normalize the weight value corresponding to each image feature.

[0130] Optionally, two convolutional neural networks can be used to determine the image features of each facial image to avoid the bias caused by a single convolutional neural network. After classifying the image features, the weight value of each facial image is predicted, and then the facial images with higher weight values ​​are exchanged, which can increase the diversity of facial images seen by each convolutional neural network.

[0131] Specifically, such as Figure 6 As shown, a batch of face images are input into the first backbone network and the second backbone network respectively. The noise label is determined according to the weight value of each face image, and the face images with higher weight values, that is, the face images with high confidence, are exchanged between the two backbone networks to increase the diversity of face images seen by each backbone network.

[0132] Step S403, sorting the weight value of each face image from large to small;

[0133] like Figure 7a As shown, the face images can be sorted and grouped according to the weight values ​​corresponding to each face image, and the constraint function L can be used to RR =max{0,δ1-(α H -α L )} makes the difference between the weight mean of a group of face images with higher weight values ​​and the weight mean of a group of face images with lower weight values ​​greater than a preset value. For example, taking the face image with the last weight value sorted as an example, after determining the target face image, it can be determined that the first label P1 corresponding to the target image is happy, and then the probability values ​​of multiple labels corresponding to the target face image (respectively P1, P2, P3, P4, P5, P6, P7) are obtained, and the maximum probability value P is determined. max , obtain the difference between the maximum probability value and the probability value corresponding to the first label (P max -P1), when the difference is greater than δ, the probability value is the largest P max The label is used as the current label of the target face image, that is, calm is used as the current label of the target face image.

[0134] Step S404: Calculate the loss value corresponding to each face image and normalize the loss value;

[0135] Step S405: Add the normalized loss value of each face image and (1-weight value) to obtain a score for each face image;

[0136] Step S406 , sorting the scores of each facial image from largest to smallest, and determining the facial images with the highest scores of 30 percent as target facial images;

[0137] Step S407: modify the first label corresponding to the target face image.

[0138] Optionally, if the difference between the maximum probability value among the probability values ​​of multiple labels corresponding to the target facial image and the probability value of the first label is greater than a preset value, the first label is changed to the label corresponding to the maximum probability value.

[0139] Optionally, facial images with larger loss values ​​may be removed from the facial images, thereby reducing the impact of uncertain facial images on the face recognition model.

[0140] For example, the loss values ​​can be 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, and the largest 30%, i.e. 10, 9, 8, can be removed, while the others remain unchanged.

[0141] The above steps can effectively repair the wrong labels corresponding to the face images, such as Figure 7b Shown are experimental data with incorrect labels fixed using the above steps.

[0142] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0143] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0144] Example 2

[0145] According to an embodiment of the present application, an image processing method is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0146] Figure 8 This is a flow chart of an image processing method according to embodiment 2 of the present application. Figure 8 As shown, the method includes the following steps:

[0147] Step S802: display the image set.

[0148] The image set includes: multiple images and a first label corresponding to each image.

[0149] In an optional embodiment, the image collection may be displayed on a device used by the user, wherein the device used by the user may be an interactive tablet, a mobile terminal, a computer or other device with a display screen, so that the user can modify the image collection.

[0150] Step S804: Determine weight values ​​of multiple images.

[0151] The weight value is used to represent the importance of the first tag.

[0152] In an optional embodiment, the displayed image set may be sent to a cloud server, and the cloud server may be used to determine weight values ​​of the multiple images.

[0153] Step S806: display the target image in the image set.

[0154] The target image is determined based on the weight values ​​of multiple images, and the first label corresponding to the target image is the label to be adjusted.

[0155] In an optional embodiment, a target image may be determined in a cloud server based on weight values ​​of multiple images, and the target image may be sent to a device used by a user so that the user can modify the first tag corresponding to the target image.

[0156] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0157] Example 3

[0158] According to an embodiment of the present application, an image processing device for implementing the above image processing method is also provided. Figure 9 As shown, the apparatus 900 includes: a receiving module 902 , a first determining module 904 , a second determining module 906 and an output module 908 .

[0159] Among them, the receiving module 902 is used to receive an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; the first determination module 904 is used to determine the weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; the second determination module 906 is used to determine the target image in the image set based on the weight values ​​of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted; the output module 908 is used to output the target image.

[0160] It should be noted that the receiving module 902, the first determination module 904, the second determination module 906, and the output module 908 correspond to steps S202 to S208 in Example 1. The examples and application scenarios implemented by the four modules and the corresponding steps are the same, but are not limited to the contents disclosed in Example 1. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.

[0161] In the above embodiment of the present application, the device further includes: an identification module, a first acquisition module and a third determination module.

[0162] Among them, the recognition module is used to identify multiple images and obtain recognition results corresponding to the multiple images; the first acquisition module is used to obtain the loss value of each image based on the recognition result and the first label corresponding to each image; the third determination module is used to determine the target image based on the weight value and / or loss value of each image.

[0163] In the above embodiment of the present application, the third determination module includes: a first processing unit and a first determination unit.

[0164] Among them, the first processing unit is used to obtain the score of each image based on the weight value and loss value of each image; the first determination unit is used to determine the target image based on the scores of multiple images.

[0165] In the above embodiment of the present application, the first processing unit includes: a normalization subunit, a first acquisition subunit and a second acquisition subunit.

[0166] Among them, the normalization subunit is used to normalize the loss value of each image to obtain the normalized result of each image; the first acquisition subunit is used to obtain the difference between the preset value and the weight value of each image to obtain the weight result of each image; the second acquisition subunit is used to obtain the sum of the normalized result and the weight result of each image to obtain the score of each image.

[0167] In the above embodiment of the present application, the first determining unit includes: a sorting subunit and a third acquiring subunit.

[0168] The sorting subunit is used to sort the multiple images in descending order of their scores; the third acquisition subunit is used to acquire a preset number of images that are ranked high to obtain the target image.

[0169] In the above embodiment of the present application, the first determination module includes: an extraction unit and a second processing unit.

[0170] Among them, the extraction unit is used to extract features from multiple images to obtain feature information corresponding to the multiple images; the second processing unit is used to process the feature information of the multiple images to obtain weight values ​​of the multiple images.

[0171] In the above embodiment of the present application, the second determination module includes: a sorting unit and an acquisition unit.

[0172] The sorting unit is used to sort the multiple images in descending order of their weight values; the acquisition unit is used to acquire a preset number of images that are ranked low to obtain a target image.

[0173] In the above embodiment of the present application, the second processing unit is further used to process feature information of multiple images using constraint conditions to obtain weight values ​​of the multiple images.

[0174] In the above embodiment of the present application, the device further includes: a second acquisition module, a fourth determination module and a first change module.

[0175] Among them, the second acquisition module is used to obtain the first recognition result of the target image, wherein the recognition result includes: multiple labels corresponding to the target image, and the probability value corresponding to each label, and the multiple labels include the first label; the fourth determination module is used to determine the second label corresponding to the maximum probability value in the recognition result; the first change module is used to change the first label corresponding to the target image to the second label.

[0176] In the above embodiment of the present application, the device further includes: a third acquisition module, a judgment module and a second modification module.

[0177] Among them, the third acquisition module is used to obtain the difference between the probability value corresponding to the second label and the probability value corresponding to the first label; the judgment module is used to judge whether the difference is greater than the preset value; the second change module is used to change the first label corresponding to the target image to the second label when the difference is greater than the preset value. In the above embodiment of the present application, the device also includes: a training module.

[0178] The training module is used to train the recognition model using the image set, wherein the loss value of the recognition model is determined based on the loss values ​​of other images in the image set except the target image.

[0179] In the above embodiment of the present application, the third acquisition module is used to obtain the first recognition result of the target image when a confirmation instruction is received, wherein the confirmation instruction is used to determine to adjust the first label.

[0180] In the above embodiment of the present application, the device further includes: a third changing module.

[0181] The receiving module is further configured to receive a third tag corresponding to the target image upon receiving a confirmation instruction; and the third changing module is configured to change the first tag corresponding to the target image into the third tag.

[0182] In the above embodiment of the present application, the device further includes: a fourth acquisition module, a fifth determination module and a fourth change module.

[0183] Among them, the receiving module is also used to receive the image selected in the target image; the fourth acquisition module is used to obtain the second recognition result of the selected image, wherein the second recognition result includes: multiple labels corresponding to the selected image, and the probability value corresponding to each label, and the multiple labels include the first label; the fifth determination module is used to determine the fourth label corresponding to the maximum probability value in the second recognition result; the fourth change module is used to change the first label corresponding to the selected image to the fourth label.

[0184] In the above embodiment of the present application, the output module includes: a second determination unit and an output unit.

[0185] The second determining unit is used to determine an output mode of the target image based on the weight value of the target image; and the output unit is used to output the target image according to the output mode of the target object.

[0186] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0187] Example 4

[0188] According to an embodiment of the present application, an image processing device for implementing the above image processing method is also provided. Figure 10 As shown, the device 1000 includes: a first display module 1002 , a determination module 1004 and a second display module 1006 .

[0189] Among them, the first display module 1002 is used to display an image collection, wherein the image collection includes: multiple images, and a first label corresponding to each image; the determination module 1004 is used to determine the weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; the second display module 1005 is used to display a target image in the image collection, wherein the target image is determined based on the weight values ​​of the multiple images, and the first label corresponding to the target image is the label to be adjusted.

[0190] It should be noted that the first display module 1002, the determination module 1004, and the second display module 1006 correspond to steps S802 to S806 in Example 2. The examples and application scenarios implemented by the three modules and the corresponding steps are the same, but are not limited to the contents disclosed in Example 2. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.

[0191] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0192] Example 5

[0193] According to an embodiment of the present application, an embodiment of a method for image processing is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0194] Figure 11 This is a flow chart of an image processing method according to Example 8 of the present application. Figure 11 As shown, the method may include the following steps:

[0195] Step S1102: Receive an image by calling a first interface.

[0196] The first interface includes: a first parameter and a second parameter, the parameter value of the first parameter is an image, and the parameter value of the second parameter is a first tag corresponding to the image.

[0197] The image in the above step may be an image in the image set selected by the user, or may be all images in the entire image set.

[0198] The first interface in the above steps can be an interface for data interaction between the cloud server and the client. The client can pass the image and the first tag corresponding to the image into the interface function as two parameters of the interface function to achieve the purpose of uploading the image to the cloud server.

[0199] Step S1104: determine the weight value of the image.

[0200] The weight value is used to represent the importance of the first tag.

[0201] Step S1106: Determine the recognition result of the image based on the weight value of the image.

[0202] The recognition result is used to indicate whether the image is a target image, and the first label corresponding to the target image is the label to be adjusted.

[0203] In an optional embodiment, an uncertainty sample recognition mechanism can be pre-deployed in the cloud server. After receiving the image uploaded by the client, the cloud server can use the pre-deployed recognition mechanism to identify the image based on the weight value of the image to determine whether the image is an uncertainty sample. If the image is an uncertainty sample, the recognition result obtained is that the image is the target image; if the image is not an uncertainty sample, the recognition result obtained is that the image is not the target image.

[0204] Step S1108: Output the recognition result by calling the second interface.

[0205] The second interface includes: a third parameter, and the parameter value of the third parameter is the identification result.

[0206] The second interface in the above steps can be an interface for data interaction between the cloud server and the client. The cloud server can pass the recognition result into the interface function as a parameter of the interface function to achieve the purpose of sending the recognition result to the client.

[0207] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0208] Example 6

[0209] According to an embodiment of the present application, an embodiment of a method for image processing is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0210] Figure 12 This is a flow chart of an image processing method according to Example 6 of the present application. Figure 12 As shown, the method may include the following steps:

[0211] Step S1202: Receive a product image set.

[0212] The product image set includes: multiple product images and a first product label corresponding to each product image.

[0213] The application scenario of the product image collection in the above steps can be an e-commerce scenario, wherein the product image can be an image of a product sold on the e-commerce platform, and the product image collection can be a collection of product images of multiple different products; the product image can be an image of a product sold in an e-commerce live broadcast, and the product image collection can be a collection of product images of multiple different products sold in an e-commerce live broadcast.

[0214] The first label in the above step can be the product information of the product, such as, but not limited to, the product name, product supplier, product type, etc. The first label can be manually labeled by a labeler.

[0215] Step S1204: Determine weight values ​​of the multiple product images.

[0216] The weight value is used to represent the importance of the first tag.

[0217] Step S1206 : determining a target product image in the product image set based on the weight values ​​of the multiple product images.

[0218] The first product label corresponding to the target product image is the label to be adjusted.

[0219] Step S1208: output the target product image.

[0220] In the above embodiment of the present application, after determining the weight values ​​of multiple product images, the method also includes: identifying multiple product images to obtain recognition results corresponding to the multiple product images; obtaining a loss value for each product image based on the recognition result and the first label corresponding to each product image; and determining the target product image based on the weight value and / or loss value of each product image.

[0221] In the above embodiment of the present application, determining the target product image based on the weight value and loss value of each product image includes: obtaining a score for each product image based on the weight value and loss value of each product image; and determining the target product image based on the scores of multiple product images.

[0222] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0223] Example 7

[0224] According to an embodiment of the present application, an image processing device for implementing the above image processing method is also provided. Figure 13 As shown, the apparatus 1300 includes: a first calling module 1302 , a first determining module 1304 , a second determining module 1306 and a second calling module 1308 .

[0225] Among them, the first calling module 1302 is used to receive an image by calling a first interface, wherein the first interface includes: a first parameter and a second parameter, the parameter value of the first parameter is the image, and the parameter value of the second parameter is the first label corresponding to the image; the first determination module 1304 is used to determine the weight value of the image, wherein the weight value is used to characterize the importance of the first label; the second determination module 1306 is used to determine the recognition result of the image based on the weight value of the image, wherein the recognition result is used to characterize whether the image is a target image, and the first label corresponding to the target image is the label to be adjusted; the second calling module 1308 is used to output the recognition result by calling a second interface, wherein the second interface includes: a third parameter, and the parameter value of the third parameter is the recognition result.

[0226] It should be noted that the first calling module 1302, the first determining module 1304, the second determining module 1306, and the second calling module 1308 correspond to steps S1102 to S1108 in Example 5. The examples and application scenarios implemented by the four modules and the corresponding steps are the same, but are not limited to the contents disclosed in Example 5. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.

[0227] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0228] Example 8

[0229] According to an embodiment of the present application, an image processing device for implementing the above image processing method is also provided. Figure 14 As shown, the apparatus 1400 includes: a receiving module 1402 , a first determining module 1404 , a second determining module 1406 and an output module 1408 .

[0230] Among them, the receiving module 1402 is used to receive a product image set, wherein the product image set includes: multiple product images, and a first product label corresponding to each product image; the first determination module 1404 is used to determine the weight values ​​of the multiple product images, wherein the weight value is used to represent the importance of the first label; the second determination module 1406 is used to determine the target product image in the product image set based on the weight values ​​of the multiple product images, wherein the first product label corresponding to the target product image is the label to be adjusted; the output module 1408 is used to output the target product image.

[0231] It should be noted that the receiving module 1402, the first determining module 1404, the second determining module 1406, and the output module 1408 correspond to steps S1202 to S1208 in Example 6. The examples and application scenarios implemented by the four modules and the corresponding steps are the same, but are not limited to the contents disclosed in Example 6. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.

[0232] In the above embodiment of the present application, the device further includes: an identification module, a first acquisition module and a third determination module.

[0233] Among them, the recognition module is used to identify multiple product images and obtain recognition results corresponding to the multiple product images; the first acquisition module is used to obtain the loss value of each product image based on the recognition result and the first label corresponding to each product image; the third determination module is used to determine the target product image based on the weight value and / or loss value of each product image.

[0234] In the above embodiment of the present application, the third determination module includes: a first processing unit and a first determination unit.

[0235] The first processing unit is used to obtain a score for each product image based on a weight value and a loss value for each product image; and the first determination unit is used to determine a target product image based on the scores of multiple product images.

[0236] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0237] Example 9

[0238] According to an embodiment of the present application, an image processing system is further provided, including:

[0239] processor; and

[0240] The memory is connected to the processor and is used to provide the processor with instructions for processing the following processing steps: receiving an image set, wherein the image set includes: multiple images and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; based on the weight values ​​of the multiple images, determining a target image in the image set, wherein the first label corresponding to the target image is the label to be adjusted; and outputting the target image.

[0241] It should be noted that the preferred implementation scheme involved in the above embodiments of this application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0242] Example 10

[0243] The embodiment of the present application can provide a computer terminal, which can be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the computer terminal can also be replaced by a terminal device such as a mobile terminal.

[0244] Optionally, in this embodiment, the computer terminal may be located in at least one network device among a plurality of network devices of a computer network.

[0245] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the image processing method: receiving an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; based on the weight values ​​of the multiple images, determining a target image in the image set, wherein the first label corresponding to the target image is the label to be adjusted; and outputting the target image.

[0246] Optionally, Figure 15 This is a structural block diagram of a computer terminal according to an embodiment of the present application. Figure 15 As shown, the computer terminal A may include: one or more (only one is shown in the figure) processors 1502 and a memory 1504.

[0247] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the image processing method and device in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned image processing method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely located relative to the processor, and these remote memories may be connected to terminal A via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0248] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: receiving an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; based on the weight values ​​of the multiple images, determining a target image in the image set, wherein the first label corresponding to the target image is the label to be adjusted; and outputting the target image.

[0249] Optionally, the processor may also execute the program code of the following steps: identifying multiple images and obtaining recognition results corresponding to the multiple images; obtaining a loss value for each image based on the recognition results and the first label corresponding to each image; and determining a target image based on the weight value and / or loss value of each image.

[0250] Optionally, the processor may also execute program code of the following steps: obtaining a score for each image based on the weight value and loss value of each image; and determining a target image based on the scores of multiple images.

[0251] Optionally, the processor may also execute the program code of the following steps: normalizing the loss value of each image to obtain the normalized result of each image; obtaining the difference between the preset value and the weight value of each image to obtain the weight result of each image; obtaining the sum of the normalized result and the weight result of each image to obtain the score of each image.

[0252] Optionally, the processor may further execute program code of the following steps: sorting the multiple images in descending order of their scores; obtaining a preset number of images with the highest scores to obtain a target image.

[0253] Optionally, the processor may further execute program codes of the following steps: extracting features from multiple images to obtain feature information corresponding to the multiple images; and processing the feature information of the multiple images to obtain weight values ​​of the multiple images.

[0254] Optionally, the processor may further execute program code of the following steps: sorting the multiple images in descending order of their weight values; obtaining a preset number of images at the bottom of the sorting order to obtain a target image.

[0255] Optionally, the processor may further execute a program code of the following steps: processing feature information of multiple images using constraint conditions to obtain weight values ​​of the multiple images.

[0256] Optionally, the processor may also execute the program code of the following steps: obtaining a first recognition result of the target image, wherein the first recognition result includes: multiple labels corresponding to the target image, and a probability value corresponding to each label, the multiple labels including the first label; determining the second label corresponding to the maximum probability value in the recognition result; and changing the first label corresponding to the target image to the second label.

[0257] Optionally, the processor may also execute the program code of the following steps: obtaining the difference between the probability value corresponding to the second label and the probability value corresponding to the first label; determining whether the difference is greater than a preset value; and if the difference is greater than the preset value, changing the first label corresponding to the target image to the second label.

[0258] Optionally, the processor may also execute the program code of the following steps: training the recognition model using the image set, wherein the loss value of the recognition model is determined based on the loss values ​​of other images in the image set except the target image.

[0259] Optionally, the processor may further execute program code of the following steps: upon receiving a confirmation instruction, obtaining a first recognition result of the target image, wherein the confirmation instruction is used to determine to adjust the first label.

[0260] Optionally, the processor may further execute program code of the following steps: upon receiving a confirmation instruction, receiving a third tag corresponding to the target image; and changing the first tag corresponding to the target image to the third tag.

[0261] Optionally, the processor may also execute the program code of the following steps: receiving a selected image from a target image; obtaining a second recognition result of the selected image, wherein the second recognition result includes: multiple labels corresponding to the selected image, and a probability value corresponding to each label, the multiple labels including a first label; determining a fourth label corresponding to the maximum probability value in the second recognition result; and changing the first label corresponding to the selected image to a fourth label.

[0262] Optionally, the processor may further execute program codes of the following steps: determining an output mode of the target image based on a weight value of the target image; and outputting the target image according to the output mode of the target object.

[0263] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: displaying an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first label; displaying a target image in the image set, wherein the target image is determined based on the weight values ​​of the multiple images, and the first label corresponding to the target image is the label to be adjusted.

[0264] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: receive the image by calling the first interface, wherein the first interface includes: a first parameter and a second parameter, the parameter value of the first parameter is the image, and the parameter value of the second parameter is the first label corresponding to the image; determine the weight value of the image, wherein the weight value is used to represent the importance of the first label; based on the weight value of the image, determine the recognition result of the image, wherein the recognition result is used to represent whether the image is a target image, and the first label corresponding to the target image is the label to be adjusted; output the recognition result by calling the second interface, wherein the second interface includes: a third parameter, and the parameter value of the third parameter is the recognition result.

[0265] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: receive a product image set, wherein the product image set includes: multiple product images, and a first product label corresponding to each product image; determine the weight values ​​of the multiple product images, wherein the weight values ​​are used to represent the importance of the first product label; based on the weight values ​​of the multiple product images, determine the target product image in the product image set, wherein the first product label corresponding to the target product image is the label to be adjusted; and output the target product image.

[0266] According to an embodiment of the present application, an image processing solution is provided. After receiving a plurality of images and a first label corresponding to each image, weight values ​​for representing the importance of the first label of the plurality of images can be determined. Then, based on the weight values ​​of the plurality of images, a target image in the image set is determined, wherein the first label corresponding to the target image is the label to be adjusted. Finally, the target image is output, thereby achieving the purpose of accurately locating the target image in the image set. It is easy to notice that after receiving the image set, a target image with a large uncertainty in the first label of the plurality of images can be determined based on the weight values ​​of the plurality of images. That is, a target image with a low degree of matching between the first label and the image can be adjusted for the first label corresponding to the target image, thereby improving the accuracy of the corresponding label of the entire image set. By training the image recognition network using the adjusted image set, an image recognition model with high recognition accuracy can be obtained, thereby achieving the technical effect of improving the recognition accuracy of the image recognition model, thereby solving the technical problem in the related art that the image recognition model obtained by training using the training samples has low recognition accuracy due to the presence of label errors and low-quality samples in the training samples.

[0267] It can be understood by those skilled in the art that Figure 15 The structure shown is for illustration only, and the computer terminal may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 15 It does not limit the structure of the above electronic device. For example, the computer terminal A may also include Figure 15 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 15 Different configurations shown.

[0268] A person skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0269] Example 15

[0270] The embodiment of the present application further provides a computer-readable storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the image processing method provided in the above embodiment.

[0271] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0272] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: receiving an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to characterize the importance of the first label; determining a target image in the image set based on the weight values ​​of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted; and outputting the target image.

[0273] Optionally, the above-mentioned storage medium is also configured to store program code for performing the following steps: identifying multiple images to obtain recognition results corresponding to the multiple images; obtaining a loss value for each image based on the recognition results and the first label corresponding to each image; and determining a target image based on the weight value and / or loss value of each image.

[0274] Optionally, the processor may also execute program code of the following steps: obtaining a score for each image based on the weight value and loss value of each image; and determining a target image based on the scores of multiple images.

[0275] Optionally, the above-mentioned storage medium is also configured to store program code for performing the following steps: normalizing the loss value of each image to obtain the normalized result of each image; obtaining the difference between the preset value and the weight value of each image to obtain the weight result of each image; obtaining the sum of the normalized result and the weight result of each image to obtain the score of each image.

[0276] Optionally, the storage medium is further configured to store program code for executing the following steps: sorting the multiple images in descending order of their scores; obtaining a preset number of images with the highest sorting scores to obtain a target image.

[0277] Optionally, the storage medium is further configured to store program codes for executing the following steps: extracting features from multiple images to obtain feature information corresponding to the multiple images; and processing the feature information of the multiple images to obtain weight values ​​of the multiple images.

[0278] Optionally, the storage medium is further configured to store program code for executing the following steps: sorting the multiple images in descending order of their weight values; obtaining a preset number of images at the bottom of the sorting to obtain a target image.

[0279] Optionally, the storage medium is further configured to store program codes for executing the following steps: processing feature information of multiple images using constraint conditions to obtain weight values ​​of multiple images

[0280] Optionally, the storage medium is further configured to store program code for executing the following steps: obtaining a first recognition result of the target image, wherein the first recognition result includes: multiple labels corresponding to the target image, and a probability value corresponding to each label, the multiple labels including the first label; determining the second label corresponding to the maximum probability value in the recognition result; and changing the first label corresponding to the target image to the second label.

[0281] Optionally, the above-mentioned storage medium is also configured to store program code for executing the following steps: obtaining the difference between the probability value corresponding to the second label and the probability value corresponding to the first label; determining whether the difference is greater than a preset value; if the difference is greater than the preset value, changing the first label corresponding to the target image to the second label.

[0282] Optionally, the above-mentioned storage medium is also configured to store program code for performing the following steps: training a recognition model using an image set, wherein the loss value of the recognition model is determined based on the loss values ​​of other images in the image set except the target image.

[0283] Optionally, the storage medium is further configured to store program code for executing the following steps: upon receiving a confirmation instruction, obtaining a first recognition result of the target image, wherein the confirmation instruction is used to determine to adjust the first label.

[0284] Optionally, the storage medium is further configured to store program codes for executing the following steps: upon receiving a confirmation instruction, receiving a third tag corresponding to the target image; and changing the first tag corresponding to the target image to the third tag.

[0285] Optionally, the storage medium is further configured to store program code for executing the following steps: receiving a selected image in a target image; obtaining a second recognition result of the selected image, wherein the second recognition result includes: multiple labels corresponding to the selected image, and a probability value corresponding to each label, the multiple labels including a first label; determining a fourth label corresponding to the maximum probability value in the second recognition result; and changing the first label corresponding to the selected image to a fourth label.

[0286] Optionally, the storage medium is further configured to store program codes for executing the following steps: determining an output mode of the target image based on a weight value of the target image; and outputting the target image according to the output mode of the target object.

[0287] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: displaying an image set, wherein the image set includes: multiple images, and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight values ​​are used to characterize the importance of the first label; displaying a target image in the image set, wherein the target image is determined based on the weight values ​​of the multiple images, and the first label corresponding to the target image is the label to be adjusted.

[0288] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: receiving an image by calling a first interface, wherein the first interface includes: a first parameter and a second parameter, the parameter value of the first parameter is the image, and the parameter value of the second parameter is the first label corresponding to the image; determining a weight value of the image, wherein the weight value is used to characterize the importance of the first label; determining a recognition result of the image based on the weight value of the image, wherein the recognition result is used to characterize whether the image is a target image, and the first label corresponding to the target image is the label to be adjusted; outputting the recognition result by calling a second interface, wherein the second interface includes: a third parameter, and the parameter value of the third parameter is the recognition result.

[0289] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: receiving a product image set, wherein the product image set includes: multiple product images, and a first product label corresponding to each product image; determining weight values ​​of the multiple product images, wherein the weight values ​​are used to characterize the importance of the first product label; determining a target product image in the product image set based on the weight values ​​of the multiple product images, wherein the first product label corresponding to the target product image is the label to be adjusted; and outputting the target product image.

[0290] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0291] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0292] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0293] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0294] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0295] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0296] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An image processing method, characterized in that: include: Receive an image set, wherein the image set includes: a plurality of images and a first label corresponding to each image; Determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first tag; Recognize the multiple images and obtain recognition results corresponding to the multiple images; Obtaining a loss value for each image based on the recognition result and the first label corresponding to each image; Obtaining a score for each image based on the weight value and loss value of each image; Based on the scores of the multiple images, a target image is output, wherein the first label corresponding to the target image is the label to be adjusted.

2. The method according to claim 1, characterized in that Determining the target image based on the scores of the multiple images includes: Sorting the multiple images in descending order of their scores; A preset number of images ranked at the top are acquired to obtain the target image.

3. The method according to claim 1, characterized in that Determining the weight values ​​of the multiple images includes: performing feature extraction on the multiple images to obtain feature information corresponding to the multiple images; The feature information of the multiple images is processed to obtain weight values ​​of the multiple images.

4. The method according to any one of claims 1 to 3, characterized in that After determining the target image in the image set based on the weight values ​​of the multiple images, the method further includes: Obtaining a first recognition result of the target image, wherein the first recognition result includes: multiple labels corresponding to the target image and a probability value corresponding to each label, wherein the multiple labels include the first label; Determine a second label corresponding to the maximum probability value in the first recognition result; The first label corresponding to the target image is changed to the second label.

5. The method according to claim 4, characterized in that Before changing the first label corresponding to the target image to the second label, the method further includes: Obtaining a difference between a probability value corresponding to the second tag and a probability value corresponding to the first tag; Determining whether the difference is greater than a preset value; If the difference is greater than the preset value, the first label corresponding to the target image is changed to the second label.

6. The method according to any one of claims 1 to 3, characterized in that After outputting the target image, the method further includes: When a confirmation instruction is received, receiving a third tag corresponding to the target image; The first label corresponding to the target image is changed to the third label.

7. The method according to any one of claims 1 to 3, characterized in that After outputting the target image, the method further includes: receiving an image selected from the target images; Obtaining a second recognition result of the selected image, wherein the second recognition result includes: multiple tags corresponding to the selected image and a probability value corresponding to each tag, wherein the multiple tags include the first tag; Determining a fourth label corresponding to the maximum probability value in the second recognition result; The first label corresponding to the selected image is changed to the fourth label.

8. The method according to any one of claims 1 to 3, characterized in that Outputting the target image includes: Determining an output mode of the target image based on the weight value of the target image; The target image is output according to the output mode of the target object.

9. The method according to claim 8, characterized in that The output mode includes one of the following: the arrangement order of the target images, and the target mark of the target image.

10. An image processing method, characterized in that: include: Displaying an image set, wherein the image set includes: a plurality of images and a first label corresponding to each image; Determining weight values ​​of the multiple images, wherein the weight values ​​are used to represent the importance of the first tag; Display a target image in the image set, wherein the target image is determined based on the scores of the multiple images, the first label corresponding to the target image is the label to be adjusted, the scores of the multiple images are determined based on the weight value and loss value of each image, the loss value of each image is determined based on the recognition result corresponding to each image and the first label, and the recognition result is obtained by recognizing the multiple images.

11. An image processing method, characterized in that: include: Receiving an image by calling a first interface, wherein the first interface includes: a first parameter and a second parameter, the parameter value of the first parameter is the image, and the parameter value of the second parameter is a first tag corresponding to the image; Determining a weight value of the image, wherein the weight value is used to represent the importance of the first tag; Determining a recognition result of the image based on the weight value of the image, wherein the recognition result is used to indicate whether the image is a target image, a first label corresponding to the target image is a label to be adjusted, the target image is determined based on the score of the image, the score of the image is determined based on the weight value and loss value of the image, the loss value of the image is determined based on the recognition result corresponding to the image and the first label, and the recognition result is obtained by recognizing the image; The recognition result is output by calling a second interface, wherein the second interface includes: a third parameter, and the parameter value of the third parameter is the recognition result.

12. An image processing method, characterized in that: include: Receive a product image set, wherein the product image set includes: a plurality of product images, and a first product label corresponding to each product image; Determining weight values ​​of the plurality of product images, wherein the weight values ​​are used to represent the importance of the first product label; Identify the multiple product images and obtain identification results corresponding to the multiple product images; Obtaining a loss value for each product image based on the recognition result and the first label corresponding to each product image; Obtaining a score for each product image based on the weight value and the loss value of each product image; Based on the scores of the multiple product images, a target product image is output, wherein the first product label corresponding to the target product image is the label to be adjusted.

13. An image processing device, characterized in that: include: A receiving module, configured to receive an image set, wherein the image set includes: a plurality of images and a first label corresponding to each image; A first determining module is configured to determine weight values ​​of the plurality of images, wherein the weight values ​​are used to represent the importance of the first tag; A recognition module, configured to recognize the plurality of images and obtain recognition results corresponding to the plurality of images; A first acquisition module, configured to obtain a loss value of each image based on the recognition result and the first label corresponding to each image; A third determination module is configured to obtain a score for each image based on the weight value and loss value of each image; An output module is configured to output a target image based on the scores of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted.

14. An image processing device, characterized in that: include: A first display module is configured to display an image set, wherein the image set includes: a plurality of images and a first label corresponding to each image; a determination module, configured to determine weight values ​​of the plurality of images, wherein the weight values ​​are used to represent the importance of the first tag; A second display module is used to display a target image in the image set, wherein the target image is determined based on the scores of the multiple images, the first label corresponding to the target image is the label to be adjusted, the scores of the multiple images are determined based on the weight value and loss value of each image, the loss value of each image is determined based on the recognition result corresponding to each image and the first label, and the recognition result is obtained by recognizing the multiple images.

15. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the image processing method according to any one of claims 1 to 12.

16. A computer terminal, characterized in that: include: A memory and a processor, wherein the processor is configured to run a program stored in the memory, wherein the image processing method according to any one of claims 1 to 12 is executed when the program is run.

17. An image processing system, characterized in that: include: processor; as well as A memory is connected to the processor and is used to provide the processor with instructions for processing the following processing steps: receiving an image set, wherein the image set includes: multiple images and a first label corresponding to each image; determining weight values ​​of the multiple images, wherein the weight value is used to characterize the importance of the first label; identifying the multiple images and obtaining recognition results corresponding to the multiple images; obtaining a loss value for each image based on the recognition result and the first label corresponding to each image; obtaining a score for each image based on the weight value and the loss value of each image; and outputting a target image based on the scores of the multiple images, wherein the first label corresponding to the target image is the label to be adjusted.

Images