Image quality detection method, device, equipment, medium and product

Abstract

Embodiments of the present disclosure provide an image quality detection method, device, equipment, medium and product. The method comprises: in response to an image detection request, obtaining a to-be-detected image corresponding to the image detection request and target inquiry information corresponding to the to-be-detected image; wherein the target inquiry information comprises information for inquiring about image quality; processing the to-be-detected image and the target inquiry information through an image detection model to obtain target feedback information; wherein the target feedback information at least comprises quality description information; the quality description information is used to describe a quality detection result of the to-be-detected image corresponding to the target inquiry information. Through interactive question and answer with the user about the quality of the to-be-detected image, based on such an interactive question and answer mode, the user can obtain more information about image quality, fine detection of the image is realized, and the interpretability and generalization ability of the image detection model are improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer processing technology, and in particular to an image quality detection method, apparatus, device, medium, and product. Background Technology

[0002] With the development of computer vision technology, information interaction through images or videos is becoming increasingly popular. For images or videos, image quality is one of the key indicators affecting their presentation. Therefore, image quality needs to be assessed in many scenarios. For example, in video evaluation, it may be necessary to assess the image quality of a video at multiple stages of the production process, including transcoding, image processing algorithms, and network transmission.

[0003] In related technologies, images or videos are usually sampled, and then the sampled images are subjectively evaluated by image quality assessment experts. This method is inefficient and requires a lot of human resources. Moreover, different image quality assessment experts may give different detection results when evaluating the same image, resulting in unstable image quality assessment results and low reliability. Summary of the Invention

[0004] This disclosure provides an image quality detection method, apparatus, device, medium, and product, which solves the technical problems of low evaluation efficiency, time-consuming and labor-intensive evaluation, and unstable image quality evaluation results caused by subjective evaluation by image quality evaluation experts. It can achieve refined image detection, improve the interpretability and generalization ability of image detection models, and allow users to obtain more information about image quality through interactive question and answer.

[0005] In a first aspect, embodiments of this disclosure provide an image quality detection method, the method comprising:

[0006] In response to an image detection request, the system acquires the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected; wherein, the target query information includes information for querying image quality.

[0007] The image to be detected and the target query information are processed by an image detection model to obtain target feedback information; wherein, the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the image to be detected corresponding to the target query information.

[0008] Secondly, embodiments of this disclosure also provide an image quality detection device, the device comprising:

[0009] An image detection request module is used to respond to an image detection request by acquiring a target image to be detected corresponding to the image detection request and target query information corresponding to the target image; wherein, the target query information includes information for querying image quality;

[0010] An image detection feedback module is used to process the image to be detected and the target query information through an image detection model to obtain target feedback information; wherein, the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the image to be detected corresponding to the target query information.

[0011] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising:

[0012] One or more processors;

[0013] Storage device for storing one or more programs.

[0014] When the one or more programs are executed by the one or more processors, the one or more processors implement the image quality detection method as described in any of the embodiments of this disclosure.

[0015] Fourthly, embodiments of this disclosure also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the image quality detection method as described in any of the embodiments of this disclosure.

[0016] Fifthly, embodiments of this disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements the image quality detection method as described in any of the embodiments of this disclosure.

[0017] The technical solution of this disclosure, in response to an image detection request, acquires the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected. Since the target query information includes information for inquiring about image quality, it supports flexible questioning of image quality-related issues, adapts to differentiated image quality detection requests, and expands the application scenarios of image detection. The image detection model processes the image to be detected and the target query information to obtain target feedback information. It can automatically generate target feedback information related to the queried image quality issues and provide rapid feedback. Since the target feedback information includes at least image quality description information, and this image quality description information describes the image quality detection result of the image to be detected corresponding to the target query information, it provides users with richer image quality detection information, enabling users to understand the image quality detection results more intuitively and accurately. By engaging in interactive question-and-answer sessions with the user regarding the image quality of the image to be detected, the interaction methods are enriched, the operation is simple and effortless, improving the efficiency of image quality detection and enhancing the image quality detection experience. Attached Figure Description

[0018] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0019] Figure 1 A schematic flowchart of an image quality detection method provided in this embodiment of the present disclosure;

[0020] Figure 2 A schematic flowchart of another image quality detection method provided in this embodiment of the present disclosure;

[0021] Figure 3 A schematic flowchart of another image quality detection method provided in this embodiment of the present disclosure;

[0022] Figure 4 This is a flowchart illustrating an optional example of a second sample image generation method for implementing the image processing method of this disclosure, provided by an embodiment of the present disclosure.

[0023] Figure 5 This is a schematic diagram of the structure of an image quality detection device provided in an embodiment of the present disclosure;

[0024] Figure 6 This is a schematic diagram of the structure of an electronic device for implementing an embodiment of the present disclosure. Detailed Implementation

[0025] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0026] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0027] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0028] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0029] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0030] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0031] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0032] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.

[0033] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0034] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0035] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0036] Figure 1 This is a flowchart illustrating an image quality detection method provided in an embodiment of this disclosure. This embodiment is applicable to scenarios involving image quality detection and interactive question-and-answer sessions. The method can be executed by an image quality detection device, which can be implemented in software and / or hardware, optionally through an electronic device such as a mobile terminal, PC, or server. Figure 1 As shown, the method in this embodiment may specifically include:

[0037] S110. In response to the image detection request, obtain the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected; wherein, the target query information includes information for querying the image quality.

[0038] The image detection request can be a user-triggered request to perform image quality detection. For example, in a user-visible image detection interface, the user can take a picture or upload an image to be detected, and can also input information about the image to be detected. After the user completes the relevant input operations, an image detection request will be generated.

[0039] In one optional embodiment of this disclosure, obtaining the image to be detected corresponding to the image detection request includes: obtaining an image set by a preset image setting control, and determining the image to be detected based on the set image. The image setting control includes an image capture control and / or an image upload control.

[0040] In this embodiment of the disclosure, the image to be detected corresponding to the image detection request can be an image captured or uploaded within the image detection interface. Alternatively, in response to the image detection request, the user-input image can be preprocessed to obtain the image to be detected corresponding to the image detection request and target query information corresponding to the image to be detected; wherein, the target query information includes information for querying image quality. For example, the size of the user-input image can be adjusted, such as cropping a portion of the image region or filling at least one edge region of the image with preset pixel values, so that the image size meets the detection criteria of the image detection model, i.e., the image to be detected is obtained.

[0041] Optionally, the target query information corresponding to the image to be detected can be the information that is input into the image to be detected. Alternatively, it can be the query information obtained after reprocessing the image and query information input by the user.

[0042] As an optional implementation of this disclosure, multiple preset image quality query messages can be displayed, an information selection operation can be received for at least one preset image quality query message, and a target query message corresponding to the image to be detected can be determined according to the information selection operation.

[0043] As another optional implementation of this disclosure, the original query information input in the image detection interface can be obtained, the target field in the original query information can be extracted, and the target field can be filled into a preset question template corresponding to the target field to obtain the target query information corresponding to the image to be detected. For example, the target query information can be "How is the quality of this image?", "Where is the image quality problem in this image?", "How can the image quality of this image be improved?", etc.

[0044] S120. The image to be detected and the target query information are processed by the image detection model to obtain target feedback information; wherein, the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the image to be detected corresponding to the target query information.

[0045] The image detection model is used to detect images and respond to target queries based on the detection results, essentially outputting target feedback information. The image detection model can be a pre-trained model. Target feedback information is used to respond to target queries. Target feedback information includes at least image quality description information, and may also include terms of address and polite expressions commonly used in interactive question-and-answer scenarios. Specifically, the image quality description information describes the image quality detection result corresponding to the target query. The image quality detection result reflects whether the image has image quality problems, and if so, details of those problems. For example, the image quality detection result may include a detailed description of the image's quality, existing image quality problems, and the location of those problems. The image quality description information can serve as target feedback information and be output to the user to respond to their queries.

[0046] For example, if a user asks, "How is the quality of this image?", the image detection model, after detecting the image to be detected, can analyze the image content to derive the image quality detection result corresponding to the target query information. To ensure the user fully understands the detected image content, the image quality description information can include a description of the image content and an image quality assessment conclusion. The image quality assessment conclusion can specifically include image quality level and / or image quality score. The image quality level can be understood as multiple pre-set levels used to assess the image quality of the image to be detected; for example, it can include at least one of levels such as very poor, poor, average, good, very good, and excellent. The image quality score can be understood as multiple scores assessing the image quality of the image to be detected, typically taken from a preset score range. The preset score range can be set according to requirements, for example, it can be 0 to 10 or 0 to 100.

[0047] It's important to note that since target feedback information is used to respond to target queries, one target query can correspond to one target feedback message. In practical applications, each time a user makes a query, a corresponding answer is provided, meaning multiple interactive question-and-answer sessions can occur with the user.

[0048] For example, a user could ask, "This image is composed of two images stitched together, the top one is the first, and the bottom one is the second. Compare the quality of the two images." After inspecting the image, the system could respond, "The first image is of better quality. The first image has clear subject texture, high overall sharpness, rich background detail, and almost no distortion caused by noise. The lighting, composition, and color of the first image are also excellent. In contrast, the second image has very poor overall sharpness, serious focus problems, noticeable noise, dull colors, a blurry subject, and a lack of texture detail. The composition and background of the second image are also poor. Therefore, the first image is of higher quality than the second image."

[0049] For example, a user can continue to ask, "After detecting the image, can you now specify the distortion type and provide their bounding box coordinates?", at which point the user can be responded with "The distortions present in the image and their locations are as follows: blurry focus (11,5), (991,997). low sharpness (0,2), (1000,987), (133,36), (951,958)".

[0050] For example, a user could continue by asking, "Please provide a brief description of the image, including specific objects and any events." The response could be, "This image seems to capture a moment of kayaking or whitewater rafting. The perspective is from a person's point of view, looking at their own legs in dark clothing, possibly a wetsuit, and wearing water boots. This person appears to be sitting in a kayak, as can be seen from the blue rim of the boat at the bottom of the image and the paddle on the left. It's an outdoor setting with a clear river or stream, and the rocky riverbed is visible through the water. On the right side of the image, a helmet is floating in the water, indicating that someone may have capsized or fallen into the water. The background shows lush green forest along the riverbank, suggesting that this water activity may be located in a natural, possibly remote, location. The overall feeling is one of action and outdoor adventure."

[0051] As an optional implementation of this disclosure, multiple queries targeting the same image to be detected can be considered as a single interaction. Multiple queries within a single transaction can be associated. When initially querying the image quality of the image to be detected, the image to be detected and the target query information can be input into the image detection model to obtain target feedback information. Subsequent queries regarding the image quality of the image to be detected can be input together with the image to be detected, the target feedback information from previous queries, and the current target query information into the image detection model for image quality detection, thereby obtaining target feedback information corresponding to the current target query information. The advantage of this approach is that it fully considers the continuity and progression of querying image quality issues, and can effectively combine the previously output image quality detection results to output more accurate target feedback information.

[0052] As an optional but non-limiting implementation, when the image to be detected has image quality problems, the target feedback information also includes an image quality feedback image; the image quality feedback image identifies the target image area with image quality problems.

[0053] Image quality issues are used to describe what problems exist in the image to be detected. For example, "low sharpness" describes that the image to be detected has low sharpness, and "blurred focus" describes that the image to be detected has a problem with blurry focus.

[0054] The image quality feedback image is an image associated with the image to be detected; specifically, it's an image that identifies regions in the image to be detected that have image quality problems. Based on the image quality feedback image, users can clearly and intuitively see the problematic target image regions in the image to be detected. These target image regions are the areas in the image to be detected that have image quality problems.

[0055] Specifically, after identifying the target image region with image quality issues, the bounding box coordinates of the target image region can be determined, and the target image region can be marked based on the bounding box coordinate information. For example, the target image region can be marked using at least one of the following methods: marking the target image region with a rectangular frame; marking the target image region with a circular frame; adding a preset marker at the center of the target image region, etc.

[0056] It should be noted that the image quality feedback image can identify at least one target region. If image quality issues exist in multiple regions of the image to be detected, then each target region can be labeled in the image quality feedback image. For example, if the image to be detected has a sharpening problem not only in the upper left corner but also in the middle right corner, then both of these problems can be labeled. The labeled image becomes the image quality feedback image and is presented to the user as target feedback information.

[0057] As an optional implementation of this disclosure, when the image to be detected includes multiple target image regions with image quality problems, the target image regions corresponding to different image quality problems can be distinguished and identified.

[0058] The technical solution of this disclosure, in response to an image detection request, acquires the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected. Since the target query information includes information for inquiring about image quality, it supports flexible questioning of image quality-related issues, adapts to differentiated image quality detection requests, and expands the application scenarios of image detection. The image detection model processes the image to be detected and the target query information to obtain target feedback information. It can automatically generate target feedback information related to the queried image quality issues and provide rapid feedback. Since the target feedback information includes at least image quality description information, and this image quality description information describes the image quality detection result of the image to be detected corresponding to the target query information, it provides users with richer image quality detection information, enabling users to understand the image quality detection results more intuitively and accurately. By engaging in interactive question-and-answer sessions with the user regarding the image quality of the image to be detected, the interaction methods are enriched, the operation is simple and effortless, improving the efficiency of image quality detection and enhancing the image quality detection experience.

[0059] Figure 2This is a flowchart illustrating another image quality detection method provided in this embodiment. Based on the above embodiments, this embodiment refines the image detection model training process. Optionally, the image detection model is trained as follows: acquiring multiple target sample images; training a deep learning model based on the target sample images and first image quality description information corresponding to the target sample images to obtain a pre-trained model; training the pre-trained model based on the target sample images and second image quality description information corresponding to the target sample images to obtain an image detection model; wherein the second image quality description information provides a higher level of detail in describing the image quality than the first image quality description information. For detailed implementation, please refer to the description of this embodiment. Technical features that are the same as or similar to those in the foregoing embodiments will not be repeated here. Figure 2 As shown, the method in this embodiment may specifically include:

[0060] S210. Obtain multiple target sample images.

[0061] In this context, the target sample image can be understood as the image data that needs to be input into the image detection model during training. The target sample image can be an image obtained through methods such as capturing a photograph, acquiring it from a preset image storage space, or generating it based on an image generation model. The target sample image can also include images obtained by reprocessing some or all of the acquired images. In short, the target sample image can include a first sample image and a second sample image. The second sample image can be obtained by reprocessing the first sample image, i.e., an augmented image generated from the first sample image. In this embodiment, multiple target sample images include target sample images of different image quality problems. The advantage of this arrangement is that it enriches the image training samples and enables the trained headshot detection model to identify different types of image quality problems.

[0062] S220. Train the deep learning model based on the target sample image and the first image quality description information corresponding to the target sample image to obtain a pre-trained model.

[0063] The first image quality description information is used to describe the image quality of the target sample image. Specifically, the first image quality description information can be generated by performing image quality detection on the target sample image using an existing image quality detection model. The existing image quality detection model can be obtained by training a machine learning model using the sample detection image and the expected image quality label corresponding to the sample detection image.

[0064] After determining the first image quality description information corresponding to each target sample image, the target sample image and its corresponding first image quality description information can be used as input data for a deep learning model to train the model and obtain a pre-trained model. For example, the deep learning model can be a deep learning model based on a visual language model framework.

[0065] As an optional but non-limiting implementation, a deep learning model is trained based on the target sample image and the first image quality description information to obtain a pre-trained model. This includes: determining the first image quality description information corresponding to each target sample image; constructing an image detection task based on the target sample image and the first image quality description information; and obtaining a visual dialogue task. The visual dialogue task includes a dialogue task associated with the image content and / or the associated scene of the image content. The deep learning model is trained through the image detection task and the visual dialogue task to obtain a pre-trained model.

[0066] The image detection task is the task that triggers the image quality detection of the target sample image. The visual dialogue task is a dialogue task that can produce a question-and-answer effect, including dialogue tasks related to image content and / or the associated scene of the image content. Visual dialogue tasks can be specifically acquired based on relevant language models. For example, if the image content is people rowing a boat, the visual dialogue task could be asking what is in the image, or asking what environment the image was likely taken in. Visual dialogue tasks can also be image dialogue tasks based on text recognition technology, i.e., tasks that require recognizing text in the image to assist in the dialogue. At least one image detection task and one visual dialogue task are required.

[0067] Specifically, after determining the first image quality description information corresponding to each target sample image, an image detection task can be constructed based on each target sample image and its corresponding first image quality description information, while simultaneously acquiring other visual dialogue tasks. Subsequently, the deep learning model is trained using these image detection and visual dialogue tasks to obtain a pre-trained model. During this training phase, the number of tasks between image detection and visual dialogue can be appropriately adjusted, enabling the pre-trained model to perform various visual dialogue and image detection tasks, achieving a balance between the model's image detection and visual question answering capabilities.

[0068] It should be noted that, in the first stage of model training in this embodiment, i.e., during the process of obtaining the pre-trained model, an image detection task is creatively introduced to train the deep learning model. This enables the obtained model to not only detect images but also to perform visual dialogue based on image content. However, although image detection capability is initially introduced in this training stage, the current image detection capability may not be accurate. For example, if the first image quality description information used to construct the image detection task has inaccurate or incomplete descriptions, the image detection capability of the pre-trained model obtained in this stage will be insufficiently accurate. Therefore, subsequent training of the pre-trained model is necessary to improve the accuracy of the model's image detection capability.

[0069] S230. The pre-trained model is trained based on the target sample image and the second image quality description information corresponding to the target sample image to obtain an image detection model; wherein, the second image quality description information describes the image quality in greater detail than the first image quality description information.

[0070] The second image quality description information describes the image quality of the target sample image. While both the first and second image quality description information describe the image quality of the target sample image, the second image quality description provides a higher level of detail than the first. In other words, compared to the first image quality description information, the second image quality description is clearer, more detailed, and more accurate. Specifically, the second image quality description information can be obtained by evaluating the target sample image based on an image quality evaluation expert knowledge base; therefore, the obtained second image quality description information can be considered to be close to the expert evaluation level.

[0071] After obtaining the second image quality description information corresponding to each target sample image, the pre-trained model can be further trained based on the target sample image and the second image quality description information corresponding to the target sample image to obtain the image detection model, thereby further improving the image detection model's ability to detect images.

[0072] As an optional but non-limiting implementation, the pre-trained model is trained based on the target sample image and the second image quality description information to obtain an image detection model, including: for a single target sample image, determining the image quality annotation information corresponding to the target sample image; converting the image quality annotation information into the second image quality description information, and converting the second image quality description information into the third image quality description information in various question-and-answer formats; and training the pre-trained model based on multiple target sample images and the third image quality description information to obtain an image detection model.

[0073] Image quality annotation information comprises annotations used to evaluate image quality from multiple perspectives. This includes the type of image quality problem, the location of the problem, and overall image evaluation metrics. Specifically, for a single target sample image, after evaluation based on an image quality evaluation expert knowledge base, various image quality annotations are obtained. These annotations are then converted into second-level image quality descriptions, and further into third-level image quality descriptions in various question-and-answer formats. Specifically, different categories of key information can be extracted from the second-level descriptions, and corresponding questions and answers can be created based on these key information. Questions and answers belonging to the same group constitute a third-level image quality description, while different groups of questions and answers result in various question-and-answer formats.

[0074] Similarly, by performing the same operation on each target sample image, the third image quality description information corresponding to each target sample image can be obtained. Furthermore, the pre-trained model can be trained based on multiple target sample images and the third image quality description information to obtain an image detection model.

[0075] It should be noted that in the second stage of model training in this embodiment, that is, in the process of continuing to train the pre-trained model to obtain the image detection model, the second image quality description information used is essentially more accurate, comprehensive and reliable training information. Therefore, by continuing to train the pre-trained model based on such second image quality description information, the image detection capability of the model can be significantly improved, that is, the image detection capability of the model can be made more accurate and reliable, and as close as possible to the expert evaluation level.

[0076] S240. In response to the image detection request, obtain the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected; wherein, the target query information includes information for querying the image quality.

[0077] S250. The image to be detected and the target query information are processed by an image detection model to obtain target feedback information; wherein, the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the image to be detected corresponding to the target query information.

[0078] The technical solution of this disclosure includes two training stages in the model training phase of the image detection model. In the first stage, the image detection model is pre-trained by creatively introducing an image detection task to train the deep learning model. This enables the model to not only detect images but also engage in visual dialogue based on image content. In the second stage of model training, more detailed and accurate second image quality description information is used to train the model, further improving the accuracy of the model's image detection capability. This solution not only performs image quality detection on the image to be detected but also allows users to ask specific questions about the image, while providing relevant feedback. Through interactive question-and-answer sessions with users regarding the image quality of the image to be detected, this solution allows users to obtain more information about image quality, achieving refined image detection and improving the interpretability and generalization ability of the image detection model.

[0079] Figure 3 This is a flowchart illustrating another image quality detection method provided in this embodiment. Based on the above embodiments, this embodiment refines the method for acquiring target sample images. Optionally, the target sample image includes a first sample image and a second sample image; acquiring multiple target sample images includes: acquiring multiple first sample images; determining multiple third sample images based on at least a portion of the first sample images; wherein the image quality of the third sample images is not lower than that of the first sample images; and performing image quality degradation processing on the multiple third sample images using various degradation processing algorithms and preset algorithm constraints to obtain multiple second sample images. Detailed implementation methods can be found in the description of this embodiment. Technical features that are the same as or similar to those in the foregoing embodiments will not be repeated here. Figure 3 As shown, the method in this embodiment may specifically include:

[0080] S310. Acquire multiple first sample images, and determine multiple third sample images based on at least a portion of the first sample images; wherein the image quality of the third sample images is not lower than that of the first sample images.

[0081] The first sample image can be an image from a real scene, and the second sample image is a sample image with image quality issues obtained by reprocessing the first sample image. Together, they constitute the target sample image.

[0082] Specifically, in the process of acquiring multiple target sample images, multiple first sample images are first acquired, and multiple third sample images are determined based on at least some of the first sample images, so that the image quality of the third sample images is not lower than that of the first sample images; that is, the determined multiple third sample images can be the images with higher image quality among the multiple first sample images, or the first sample images can be optimized to obtain multiple third sample images.

[0083] Optionally, multiple third sample images can be selected from multiple first sample images. Specifically, this can be done by determining whether the image quality score of each first sample image exceeds a scoring threshold. If it is determined that the score exceeds the threshold, the corresponding first sample image is used as the third sample image. The image quality score can be based on three dimensions: image noise, blur, and blockiness. Alternatively, more dimensions can be set based on actual needs.

[0084] To improve the accuracy of image detection models, the target sample images can be made as close as possible to the real-world scene. For example, the types of image quality problems covered by the target sample images can be made similar to those in the real-world scene; the combinations and patterns of image quality problems in the target sample images can be made similar to those in the real-world scene; and the distribution of image quality problems in the target sample images can be made similar to those in the real-world scene. In other words, the target sample images should be made as close as possible to the real-world scene in all aspects so that the trained image detection model can match the image detection needs in real-world scenarios.

[0085] As an optional but non-limiting implementation, obtaining multiple first sample images includes: obtaining multiple candidate images; for each candidate image, determining the image feature values ​​of the candidate image under multiple feature extraction dimensions; and for each feature extraction dimension, sampling the multiple candidate images based on the image feature values ​​of the multiple candidate images under the feature extraction dimension to obtain multiple first sample images.

[0086] The candidate images are the most original images obtained, and can be obtained from multiple candidate images based on open-source image data. Feature extraction dimensions may include at least one of the following: color richness, image quality score, landscape / portrait orientation, sharpness, contrast, image entropy, and brightness.

[0087] Specifically, for each candidate image, its image feature value under each of the aforementioned feature extraction dimensions can be determined. This process is repeated to obtain the image feature values ​​for all candidate images under each feature extraction dimension. Furthermore, for a specific feature extraction dimension, multiple candidate images can be sampled based on their image feature values ​​under that dimension to select images that satisfy a uniform distribution within that dimension. This process is repeated for each feature extraction dimension to select images that satisfy a uniform distribution across all dimensions. After obtaining multiple images that meet the requirements, duplicate images can be removed or adjusted to obtain multiple first sample images.

[0088] Here, we can use an example to illustrate the detailed principle of obtaining the first sample image. For instance, suppose we obtain 10,000 candidate images, and there are eight feature extraction dimensions, specifically color richness, image quality score, landscape / portrait orientation, sharpness, contrast, image entropy, and brightness. First, for a candidate image 'a', we can determine the image feature values ​​of candidate image 'a' under the above eight feature extraction dimensions; similarly, we can determine the image feature values ​​of each of the 1,000 candidate images under the above eight feature extraction dimensions. Next, regarding one of the eight feature extraction dimensions mentioned above, such as color richness, its image feature value is the color richness value, and the range corresponding to the color richness value is [0,1]. Assuming this image feature value range can be divided into four sub-ranges: [0,0.25), [0.25,0.5), [0.5,0.75), and [0.75,1], when sampling multiple candidate images under this color richness feature extraction dimension, the color richness values ​​of the sampled images should be evenly distributed within these four sub-ranges. If 6000 images are sampled under the color richness feature extraction dimension, then the color richness values ​​of these 6000 images are evenly distributed within the four sub-ranges. That is, the difference in the number of images corresponding to the color richness values ​​of these 6000 images within the ranges [0,0.25), [0.25,0.5), [0.5,0.75), and [0.75,1] is less than a preset difference, indicating a uniform distribution. Similarly, sampling based on the same principle is performed on other feature extraction dimensions to obtain images that satisfy a uniform distribution across each feature extraction dimension. After obtaining all the sampled and filtered images, duplicate images can be removed, and the image feature values ​​of some feature dimensions can be adjusted. After adjustment, further removal is performed to obtain the final multiple first sample images. The number of removal and adjustment operations and the degree of processing need to be determined based on the distribution of the retained first sample images, until the final multiple retained first sample images not only cover the image feature value ranges under each different feature extraction dimension, but also the number of images corresponding to each image feature value range is evenly distributed. For example, if the final processed first sample images are 2000, then the image feature values ​​corresponding to these 2000 first sample images not only cover the image feature value ranges under the above eight feature extraction dimensions, but also the number of images corresponding to each image feature value range is evenly distributed with little difference.

[0089] It should be noted that the above is merely an example of uniform sampling across multiple feature value ranges in multiple feature extraction dimensions, and is not a limitation. In this embodiment, uniform sampling of multiple candidate images based on different feature extraction dimensions ensures that the first sample image obtained after sampling can cover more comprehensive and uniform feature information. This facilitates the subsequent model's access to more comprehensive training data during training, thereby helping to improve the model's stability and generalization ability.

[0090] S320. Multiple degradation processing algorithms and preset algorithm constraints are used to perform image quality degradation processing on multiple third sample images to obtain multiple second sample images.

[0091] The degradation processing algorithm is used to degrade the image, resulting in a certain degree of image quality reduction compared to the original image. Different degradation processing algorithms can produce different degradation effects. For example, using a focus blur degradation processing algorithm can cause the degraded image to exhibit a blurring phenomenon caused by lens misfocusing. Algorithm constraints are conditions used to constrain the various degradation processing algorithms used during the degradation processing process. For example, degradation processing algorithms that can achieve motion blur effects can only be applied to moving objects (such as people or vehicles), and cannot be used on stationary objects. Another example is that for the same object, a degradation processing algorithm that can cause overexposure cannot be used simultaneously with one that can cause underexposure. Such restrictions serve as algorithm constraints to constrain the degradation processing algorithms used during the degradation process, ensuring the degradation processing is reasonable and the resulting multiple second sample images more closely reflect the actual objective situation.

[0092] The degradation processing algorithm and its constraints can be pre-set based on actual needs. Once determined in advance, multiple degradation processing algorithms and preset constraints can be used to degrade the image quality of multiple third sample images to obtain multiple second sample images.

[0093] As an optional but non-limiting implementation, the algorithm constraints shall include at least the constraints on the applicability of the degradation processing algorithm to the object processed by the algorithm and / or the constraints on the superposition of multiple degradation processing algorithms.

[0094] In this context, the processing object refers to the object in the image to be detected that needs to be degraded. The processing object is not limited to specific objects in the image; anything in the image that requires processing can be considered a processing object. For example, the processing object can include not only faces, animals, and text in the image, but also edge regions.

[0095] Specifically, different degradation processing algorithms are suitable for different objects, and the same degradation processing algorithm may not be suitable for other objects. For example, degradation processing algorithms that can achieve motion blur effects can only be applied to moving objects (such as people, vehicles, or animals), but not to stationary objects (such as buildings). As another example, degradation processing algorithms that can achieve excessive facial beautification effects can only be applied to facial areas, but not to images that do not contain facial areas (such as landscape images). There are various application requirements, and the applicability relationship between degradation processing algorithms and the objects they process can be used as a constraint.

[0096] Furthermore, when multiple degradation algorithms are applied to the same image in combination, the effects may cancel each other out. For example, if an algorithm that produces oversaturation is applied to an image, and then an algorithm that produces undersaturation is applied, the two saturation effects will cancel each other out. Similarly, if an algorithm that produces overexposure is applied to an image, and then an algorithm that produces underexposure is applied, the two exposure effects will cancel each other out. It is evident that when multiple degradation algorithms are used in combination, some may produce contradictory effects. These contradictory algorithms should not be used simultaneously, as their combined use will cancel each other out, failing to achieve the desired image degradation. Therefore, the way these multiple degradation algorithms are used in combination can serve as a constraint.

[0097] Therefore, constraints on the applicability of degradation processing algorithms to the objects they process and / or the overlapping use of multiple degradation processing algorithms can be used as constraints. In addition, algorithm constraints can also include limits on the number of degradation processing algorithms that can be overlapped. The specific algorithm constraints can be differentiated based on actual needs and are not detailed here. The aim is to limit the conditions under which degradation processing algorithms are used, thereby improving the rationality and reliability of image degradation processing and making the degradation effect more realistic and reliable.

[0098] As an optional but non-limiting implementation, at least one of the following degradation processing algorithms can be used in the embodiments of this disclosure to address different image quality problems:

[0099] Algorithms for compressing distortion and blockiness: Convert the image into a video at a preset frame rate (e.g., 1 frame / second), and then transcode the video. This can simulate the blockiness artifacts introduced by the image during the compression and transcoding process.

[0100] An algorithm for adding jagged edges works as follows: First, edge detection is performed on the image to obtain a mask image of the edge region. Then, the entire image is compressed, multiplied with the mask image of the edge region, and the distortion of the edge parts is preserved. Finally, this is added to the original image to simulate the jagged edge effect introduced by the image processing algorithm.

[0101] An algorithm for adding edge ringing effects works by first downsampling the image to half the original resolution, then performing JPEG compression on the entire image, and finally upsampling it back to the original resolution. This can simulate edge artifacts introduced by image processing algorithms.

[0102] The algorithm used to achieve the edge blurring effect is as follows: first, edge detection is performed on the image, and then Gaussian blur is added to the edge area of ​​the image to simulate the edge blurring problem introduced by the preprocessing algorithm.

[0103] The algorithm used to achieve the edge ripple effect works as follows: First, edge detection is performed on the image to obtain a mask image of the edge region. The image is then downsampled to half the resolution of the original image, and then upsampled back to the original resolution. Due to the quantization of pixel values ​​in the sampling strategy, a stepped jagged effect will appear at the edge of the lines. After multiplying with the edge mask image, the distortion of the edge part is preserved, and then added to the original image to simulate the edge ripple effect introduced by the image processing algorithm.

[0104] The algorithm used to achieve the focus blur effect is to add a Gaussian blur to simulate the blurring caused by lens misfocusing.

[0105] Algorithm for adding noise: Gaussian or Poisson noise is added to the image to simulate image noise generated by the shooting device in a low-light shooting environment.

[0106] The algorithm used to achieve motion blur effect is to generate a blur kernel at a specific angle and superimpose it onto the original image, thereby simulating the motion blur phenomenon caused by the movement of objects.

[0107] The algorithm used to achieve the overexposure effect calculates the highest brightness value of the image and increases it to simulate the overexposure effect of an image in a strong light shooting scene.

[0108] The algorithm used to add an underexposure effect calculates the minimum brightness value of the image and adjusts it to be reduced, thereby simulating the underexposure effect of an image in a low-light shooting scene.

[0109] The algorithm used to achieve the overly dark effect is to darken the entire image by using gamma correction.

[0110] The algorithm used to achieve the oversharpening effect is as follows: an oversharpening filter is created, and the original image is sharpened multiple times to produce the oversharpening effect.

[0111] The algorithm used to achieve the oversaturation effect is to convert the image from the BGR (Blue, Green, Red) color space to the HSV (Hue, Saturation, Value) color space, and then increase the saturation value to simulate the effect of overly vivid colors in the image.

[0112] The algorithm used to achieve the undersaturation effect is to convert the image from the BGR color space to the HSV color space and adjust the saturation value by a small amount to simulate the effect of the image colors becoming too dark.

[0113] The algorithm used to achieve the interlaced scanning effect simulates the visual flickering problem caused by the high-frequency alternation of odd and even rows in early television signal transmission by randomly deleting pixel values ​​from odd or even rows of the image.

[0114] An algorithm for achieving excessive facial beautification effects works by first detecting faces in an image to obtain a mask image of the face region, then overlaying one or more beautification algorithms onto the entire image and multiplying it with the mask image of the face region to simulate the excessive beautification effect introduced by the beautification algorithm.

[0115] The algorithm used to achieve the smearing effect in images is as follows: text detection is performed on the image, and a Gaussian blur effect is applied to the text area to simulate the smearing effect of the text area.

[0116] As an optional but non-limiting implementation, multiple degradation processing algorithms and preset algorithm constraints are used to perform image quality degradation processing on multiple third sample images to obtain multiple second sample images. This includes: labeling each first sample image to determine the image quality labeling information corresponding to each first sample image; wherein the image quality labeling information includes the problem type corresponding to the image quality problem in the first sample image; determining the first image distribution information of the multiple first sample images under a preset degradation type based on the image quality problem and problem type, and determining the second image distribution information corresponding to the target sample image based on the preset degradation type and the first image distribution information; and using multiple degradation processing algorithms and preset algorithm constraints to perform image quality degradation processing on the third sample images based on the second image distribution information to obtain multiple second sample images.

[0117] The "problem type" reflects the type of image quality problem present in the first sample image. Problem types include at least one of the following: meaningless solid color, jagged edges, low sharpness, too dark, compression distortion block effect, focus blur, overexposure, noise, motion blur, underexposure, interlaced scanning, ringing effect, moiré pattern, banding, oversaturation, undersaturation, over-sharpness, blurred edges, edge ripples, excessive facial beautification, and image smearing. It should be noted that more problem types can be set based on actual needs. Image quality annotation information includes the problem type corresponding to the image quality problem in the first sample image; the image quality annotation information may also include the location area of ​​the image quality problem and the overall image evaluation index.

[0118] Specifically, for each first sample image, image quality issues can be labeled, which can be achieved by adding image quality issue labels. When labeling a first sample image, all image quality issues present in the image are labeled. For example, if there are two types of image quality issues in the first sample image, then two image quality issue labels are added to the first sample image.

[0119] The preset degradation type is determined according to preset rules. Optionally, the preset degradation type can be determined according to the image quality problem's presentation area in the first sample image; and / or according to the problem type corresponding to the image quality problem. Specifically, if the image quality problem is divided according to its presentation area in the first sample image, the determined preset degradation types include local degradation, global degradation, and both local and global degradation. Local degradation means that the image quality problem only appears in a local area of ​​each first sample image; global degradation means that the image quality problem only appears in the global area of ​​each first sample image; both local and global degradation means that the image quality problem appears in both a local area and the global area of ​​the first sample image. In other words, for a certain image quality problem, the image quality problem may appear in the global area of ​​some images or in a local area of ​​one image.

[0120] If we categorize the image quality issues according to their corresponding problem types, the determined preset degradation type can be a combination of image quality issues. For example, it could be "compression distortion block effect" and "low sharpness", or "motion blur" and "low sharpness".

[0121] It's important to note that the preset degradation type determined based on the image quality issues in the first sample image essentially reflects the distribution of each individual image quality issue within the real-world scene, indicating where a particular image quality issue might be located. Conversely, the preset degradation type determined based on the specific issue type within the image quality issue reflects which image quality issues might combine in the real-world scene. Therefore, the preset degradation type determined based on these two dimensions is more consistent with the image conditions in real-world scenes.

[0122] After determining the preset degradation type, the first image distribution information can be determined based on the ratio between the number of first sample images belonging to the preset degradation type and the total number of first sample images. Then, the second image distribution information corresponding to the target sample image can be determined based on the preset degradation type and the first image distribution information. Specifically, the first image distribution information can be directly used as the second image distribution information, or the first image distribution information can be converted into relative image distribution information among multiple preset degradation types, and then the relative image distribution information can be used as the second image distribution information. For example, if the first image distribution information is such that the ratio between local degradation, global degradation, and both local and global degradation is 1.2:1:0.9, this ratio can be directly used as the second image distribution information; alternatively, based on this ratio, relative image distribution information can be derived where "local degradation is 20% greater than global degradation, and both local and global degradation are 10% less than global degradation," and this relative image distribution information can be used as the second image distribution information.

[0123] It should be noted that the first image distribution information in this embodiment essentially reflects the actual distribution of various preset degradation types in a real scene image. The second image distribution information essentially reflects the distribution of various preset degradation types that should also be present in the target sample image. This allows the target sample image to realistically simulate the image conditions of a real scene. Furthermore, after determining the second image distribution information, it can provide guidance for the process of degrading the third sample image to obtain the second sample image. That is, it can guide the process of degrading the third sample image to determine the number of third sample images to be degraded and the type of degradation treatment to be applied to the third sample images.

[0124] After obtaining the second image distribution information, the third sample image can be processed using various degradation processing algorithms and preset algorithm constraints based on the second image distribution information to obtain multiple second sample images. The image degradation processing includes global degradation only, local degradation only, and both local and global degradation.

[0125] Specifically, the global degradation processing includes: performing global image quality degradation processing on the third sample image according to the degradation processing algorithm to obtain a globally degraded image. The obtained globally degraded image can then be used as the second sample image.

[0126] The local degradation processing includes: detecting the third sample image based on a preset local detection method to determine the local degradation area and the corresponding mask image; performing global image quality degradation processing on the third sample image based on a degradation processing algorithm to obtain a globally degraded image; and fusing the mask image and the globally degraded image to obtain a locally degraded image. The obtained locally degraded image can also be used as a second sample image. The preset local detection method includes segmentation-based local detection, object-based local detection, and edge-based local detection.

[0127] Among them, segmentation-based local detection methods refer to decomposing objects in an image using GPT or LLaVa, inputting the object segmentation algorithm (such as the Segment Anything algorithm) to obtain the segmented region of the object. Object-based local detection methods refer to using different detection methods for different objects. For example, to detect a face in an image, object-based local detection methods can be used to obtain the face region in the image. Edge-based local detection methods refer to using edge detection algorithms (such as the Canny edge detection algorithm) to detect the lines in the image, thus obtaining the region where the lines are located. By classifying the preset local detection methods, it is easier to quickly and accurately identify the degradation processing algorithm to be used in subsequent degradation processing, improving the efficiency of image degradation processing.

[0128] Local and global degradation processing includes: first performing global image quality degradation processing on the third sample image to obtain a globally degraded image, and then performing local image quality degradation processing on the globally degraded image; or, first performing local image quality degradation processing on the third sample image to obtain a locally degraded image, and then performing global image quality degradation processing on the locally degraded image.

[0129] It should be noted that in this embodiment, the first sample image reflects the image of the real scene, while the second sample image reflects the image data after degradation processing. Specifically, in obtaining the second sample image, based on the distribution of the real scene and the constraints of the degradation process, multiple sample images undergo relatively realistic and reasonable degradation processing. This makes the resulting multiple second sample images more closely resemble the real scene. Therefore, the target sample image composed of the first and second sample images can realistically simulate the real scene, thereby helping to improve the accuracy and generalization ability of the image detection model.

[0130] S330. Using the first sample image and the second sample image as target sample images, train the deep learning model based on the target sample images and the first image quality description information corresponding to the target sample images to obtain a pre-trained model.

[0131] S340. The pre-trained model is trained based on the target sample image and the second image quality description information corresponding to the target sample image to obtain an image detection model; wherein, the second image quality description information describes the image quality in greater detail than the first image quality description information.

[0132] S350, In response to the image detection request, obtain the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected; wherein, the target query information includes information for querying the image quality.

[0133] S360. The image to be detected and the target query information are processed by the image detection model to obtain target feedback information; wherein, the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the image to be detected corresponding to the target query information.

[0134] The technical solution of this disclosure, in the process of training the image detection model, firstly, in the training data preparation stage, the first sample image obtained based on the candidate image sampling can cover the feature value range of various image under different feature extraction dimensions; multiple second sample images are obtained by using multiple degradation processing algorithms and preset algorithm constraints to perform image quality degradation processing on multiple third sample images (determined from the first sample images), which can reasonably and realistically simulate the relevant degradation images. The target sample image composed of the first sample image and the second sample image can realistically, reasonably and appropriately simulate the image situation in the real scene, which is beneficial to improving the generalization ability and accuracy of the image detection model.

[0135] Figure 4 This is a flowchart illustrating an optional example of a second sample image generation method for implementing the image processing method of this disclosure, as provided in an embodiment of the present disclosure. Figure 4 As shown, the process of generating the second sample image in this example may specifically include:

[0136] Step 1: Dataset Sampling. Obtain n candidate images, including candidate image 1, candidate image 2, ..., candidate image n. Then, calculate the feature values ​​of each candidate image under m feature extraction dimensions. Then, sample the candidate images based on the feature values ​​of each candidate image under feature extraction dimension 1, feature extraction dimension 2, ..., feature extraction dimension m to obtain multiple first sample images evenly distributed in each feature extraction dimension and in different feature value intervals of each feature extraction dimension.

[0137] Step 2: Data Annotation. Based on the image quality evaluation expert knowledge base, the first sample image is annotated according to the problem type corresponding to the image quality issue, thereby obtaining the image quality annotation information corresponding to the first sample image.

[0138] Step 3: Analyze the distribution of image quality issues. Specifically, analyze the distribution of combinations of image quality issues and the distribution of areas where these issues are present. Combinations of image quality issues include "compression distortion" + "low sharpness" and "motion blur" + "low sharpness," etc. Areas where image quality issues are present can include localized degradation, global degradation, and degradation in both local and global areas.

[0139] Step 4: Set the degradation processing strategy. Degradation processing strategies can include limiting the effect of specific image quality issues to specific objects and eliminating the problem of conflicting image quality issues. For example, limiting the effect of specific image quality issues to specific objects makes the processed degraded image more suitable for the actual application scenario. For instance, a degradation processing algorithm that achieves motion blur can only be applied to moving objects (such as people, vehicles, or animals), but not to stationary objects (such as buildings). Another example is a degradation processing algorithm that achieves excessive facial beautification effects, which can only be applied to facial areas, but not to images that do not contain facial areas. Eliminating the problem of conflicting image quality issues can exclude combinations of image quality issues that would not occur simultaneously in a real-world scenario. For example, if a degradation processing algorithm that produces an oversaturation effect is applied to an image, and then a degradation processing algorithm that produces an undersaturation effect is applied, the two saturation effects will cancel each other out. Similarly, if a degradation processing algorithm that produces an overexposure effect is applied to an image, and then a degradation processing algorithm that produces an underexposure effect is applied, the two exposure effects will cancel each other out. It is evident that when multiple degradation processing algorithms are used in combination, some degradation processing algorithms may produce contradictory effects. These algorithms that produce contradictory effects should not be used simultaneously.

[0140] Step 5: Degradation Processing. In this step, the first sample image can be processed to obtain a high-quality image, i.e., the third sample image, whose image quality meets the preset quality conditions. Then, the third sample image is degraded to obtain the second sample image. Specifically, the third sample image can be globally degraded using multiple degradation algorithms or a combination of multiple degradation algorithms to obtain the second sample image; it can also be locally degraded using multiple degradation algorithms or a combination of multiple degradation algorithms to obtain a locally degraded image, and then globally degraded using multiple degradation algorithms or a combination of multiple degradation algorithms to obtain the second sample image; or, the third sample image can be globally degraded using multiple degradation algorithms or a combination of multiple degradation algorithms to obtain a globally degraded image, and then locally degraded using multiple degradation algorithms or a combination of multiple degradation algorithms to obtain the second sample image. In local degradation processing, the target object in the third sample image is first determined, and a generated object mask image is determined. Then, a degradation algorithm is used to degrade the third sample image to obtain a globally degraded image. The third sample image, the globally degraded image, and the object mask image are then fused to obtain a locally degraded image. Specifically, the globally degraded image is multiplied by the object mask image to obtain a degraded region image, which is then fused with the third sample image to obtain the locally degraded image. Determining the target object in the third sample image and determining the generated object mask image can specifically include: segmenting, detecting, or edge-detecting the target object in the third sample image to obtain the object image region of the target object in the third sample image, and determining the object mask image of the target object based on the object image region. The target object can include target objects, faces, text, or edge information in the third sample image. In this step, the third sample image can be degraded according to the distribution of image quality problems and the degradation processing strategy.

[0141] By adopting this technical solution, the training sample set composed of the obtained second sample image and the first sample image can cover the actual distribution of image quality problems and fitting quality problems that occur in real-world scenarios as comprehensively as possible. The image detection model obtained by training the deep learning model through this training sample set has better robustness, higher accuracy and stronger generalization ability when identifying image quality problems.

[0142] Figure 5 This is a schematic diagram of the structure of an image quality detection device provided in an embodiment of the present disclosure, as shown below. Figure 5As shown, the device includes an image detection request module 510 and an image detection feedback module 520. The image detection request module 510 is used to, in response to an image detection request, acquire a target image to be detected corresponding to the image detection request and target query information corresponding to the target image; wherein the target query information includes information for inquiring about image quality. The image detection feedback module 520 is used to process the target image to be detected and the target query information through an image detection model to obtain target feedback information; wherein the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the target image corresponding to the target query information.

[0143] The technical solution of this embodiment responds to an image detection request by an image detection request module 510, acquiring the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected. Since the target query information includes information for inquiring about image quality, it supports flexible questioning of image quality-related issues, adapting to differentiated image quality detection requests and expanding the application scenarios of image detection. An image detection feedback module 520 processes the image to be detected and the target query information through an image detection model to obtain target feedback information. It can automatically generate target feedback information for the inquired image quality issues through the image detection model and provide rapid feedback. Since the target feedback information includes at least image quality description information, and the image quality description information describes the image quality detection result of the image to be detected corresponding to the target query information, it provides users with richer image quality detection information, allowing users to understand the image quality detection results more intuitively and accurately. By engaging in interactive question-and-answer sessions with the user regarding the image quality of the image to be detected, the interaction methods are enriched, the operation is simple and effortless, improving the efficiency of image quality detection and enhancing the image quality detection experience.

[0144] Based on any optional technical solution in the embodiments of this disclosure, the image quality detection device further includes: an image detection model training module, used to train the image detection model.

[0145] Based on any optional technical solution in the embodiments of this disclosure, the image detection model training module may include: a target sample image acquisition submodule, a first-stage training submodule, and a second-stage training submodule. The target sample image acquisition submodule is used to acquire multiple target sample images; the first-stage training submodule is used to train a deep learning model based on the target sample images and first image quality description information corresponding to the target sample images to obtain a pre-trained model; the second-stage training submodule is used to train the pre-trained model based on the target sample images and second image quality description information corresponding to the target sample images to obtain an image detection model; wherein the second image quality description information provides a higher level of detail in describing the image quality than the first image quality description information.

[0146] Based on any optional technical solution in the embodiments of this disclosure, the first-stage training submodule may include: a training task determination unit and a first-stage training unit. The training task determination unit is configured to determine first image quality description information corresponding to each of the target sample images, construct an image detection task based on the target sample images and the first image quality description information, and obtain a visual dialogue task; wherein the visual dialogue task includes a dialogue task associated with image content and / or the associated scene of the image content; the first-stage training unit is configured to train the deep learning model through the image detection task and the visual dialogue task to obtain a pre-trained model.

[0147] Based on any optional technical solution in the embodiments of this disclosure, the second-stage training submodule may include: an image quality annotation information determination unit, an image quality description information conversion unit, and a second-stage training unit. The image quality annotation information determination unit is used to determine image quality annotation information corresponding to a single target sample image; the image quality description information conversion unit is used to convert the image quality annotation information into second image quality description information, and then convert the second image quality description information into third image quality description information in various question-and-answer formats; the second-stage training unit is used to train the pre-trained model based on multiple target sample images and the third image quality description information to obtain an image detection model.

[0148] Based on any optional technical solution in the embodiments of this disclosure, the target sample image may include a first sample image and a second sample image. Further, the target sample image acquisition submodule may include: a sample image determination unit and an image quality degradation processing unit. The sample image determination unit is used to acquire multiple first sample images and determine multiple third sample images based on at least a portion of the first sample images; wherein the image quality of the third sample images is not lower than that of the first sample images; and the image quality degradation processing unit is used to perform image quality degradation processing on the multiple third sample images using multiple degradation processing algorithms and preset algorithm constraints to obtain multiple second sample images.

[0149] Based on any optional technical solution in the embodiments of this disclosure, the sample image determination unit may include: an image feature value determination subunit and an image sampling subunit. The image feature value determination subunit is used to acquire multiple candidate images, and for each candidate image, determine the image feature values ​​of the candidate image under multiple feature extraction dimensions. The image sampling subunit is used to sample the multiple candidate images for each feature extraction dimension based on the image feature values ​​of the multiple candidate images under the feature extraction dimension, to obtain multiple first sample images.

[0150] Based on any optional technical solution in the embodiments of this disclosure, the image quality degradation processing unit may include: an image annotation subunit, an image distribution information determination subunit, and an image quality degradation processing subunit. The image annotation subunit is used to annotate each of the first sample images to determine the image quality annotation information corresponding to each of the first sample images; wherein the image quality annotation information includes the problem type corresponding to the image quality problem in the first sample image; the image distribution information determination subunit is used to determine the first image distribution information of multiple first sample images under a preset degradation type based on the image quality problem and the problem type, and to determine the second image distribution information corresponding to the target sample image based on the preset degradation type and the first image distribution information; the image quality degradation processing subunit is used to perform image quality degradation processing on the third sample image based on the second image distribution information, using multiple degradation processing algorithms and preset algorithm constraints, to obtain multiple second sample images.

[0151] Based on any optional technical solution in the embodiments of this disclosure, the algorithm constraints include at least constraints on the applicability relationship between the degradation processing algorithm and the object processed by the algorithm and / or constraints on the superposition of multiple degradation processing algorithms.

[0152] Based on any optional technical solution in the embodiments of this disclosure, if the image to be detected has an image quality problem, the target feedback information further includes an image quality feedback image; the image quality feedback image identifies the target image area where the image quality problem exists.

[0153] The image quality detection device provided in this disclosure can execute the image quality detection method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects for executing the image quality detection method.

[0154] It is worth noting that the various units and modules included in the above-mentioned device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this disclosure.

[0155] The following is for reference. Figure 6 The diagram illustrates a structural schematic of an electronic device (e.g., a terminal device or a server) 600 suitable for implementing embodiments of the present disclosure. The terminal device in the embodiments of the present disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0156] like Figure 6 As shown, electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 601, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 602 or a program loaded from storage device 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of electronic device 600. Processing device 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0157] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic device 600 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 An electronic device 600 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0158] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, it performs the functions defined in the methods of embodiments of this disclosure.

[0159] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0160] The electronic device provided in this disclosure and the image quality detection method provided in the above embodiments belong to the same inventive concept. Technical details not described in detail in this disclosure can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0161] This disclosure provides a computer storage medium storing a computer program that, when executed by a processor, implements the image quality detection method provided in the above embodiments.

[0162] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0163] According to one or more embodiments of this disclosure, [Example 1] provides an image quality detection method, comprising: in response to an image detection request, acquiring a target image to be detected corresponding to the image detection request and target query information corresponding to the target image; wherein the target query information includes information for querying image quality; processing the target image to be detected and the target query information through an image detection model to obtain target feedback information; wherein the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the target image corresponding to the target query information.

[0164] According to one or more embodiments of this disclosure, [Example 2] provides the method of Example 1, further comprising: optionally, the image detection model is trained by: acquiring multiple target sample images; training a deep learning model based on the target sample images and first image quality description information corresponding to the target sample images to obtain a pre-trained model; training the pre-trained model based on the target sample images and second image quality description information corresponding to the target sample images to obtain an image detection model; wherein the second image quality description information provides a higher level of detail in describing the image quality than the first image quality description information.

[0165] According to one or more embodiments of this disclosure, Example 3 provides the method of Example 2, which further includes: optionally, training the deep learning model based on the target sample image and the first image quality description information to obtain a pre-trained model includes: determining the first image quality description information corresponding to each of the target sample images, constructing an image detection task based on the target sample image and the first image quality description information, and obtaining a visual dialogue task; wherein the visual dialogue task includes a dialogue task associated with image content and / or the associated scene of the image content; training the deep learning model through the image detection task and the visual dialogue task to obtain a pre-trained model.

[0166] According to one or more embodiments of this disclosure, Example 4 provides the method of Example 2, which further includes: optionally, training the pre-trained model based on the target sample image and the second image quality description information to obtain an image detection model includes: for a single target sample image, determining image quality annotation information corresponding to the target sample image; converting the image quality annotation information into second image quality description information, and converting the second image quality description information into third image quality description information in multiple question-and-answer formats; training the pre-trained model based on multiple target sample images and the third image quality description information to obtain an image detection model.

[0167] According to one or more embodiments of this disclosure, Example 5 provides the method of Example 2, which further includes: optionally, the target sample image includes a first sample image and a second sample image; the acquisition of multiple target sample images includes: acquiring multiple first sample images, determining multiple third sample images based on at least a portion of the first sample images; wherein the image quality of the third sample images is not lower than the image quality of the first sample images; and performing image quality degradation processing on the multiple third sample images using multiple degradation processing algorithms and preset algorithm constraints to obtain multiple second sample images.

[0168] According to one or more embodiments of this disclosure, Example Six provides the method of Example Five, which further includes: optionally, obtaining multiple first sample images includes: obtaining multiple candidate images; for each candidate image, determining the image feature value of the candidate image under multiple feature extraction dimensions; for each feature extraction dimension, sampling the multiple candidate images according to the image feature values ​​of the multiple candidate images under the feature extraction dimension to obtain multiple first sample images.

[0169] According to one or more embodiments of this disclosure, Example 7 provides the method of Example 5, which further includes: Optionally, the step of using multiple degradation processing algorithms and preset algorithm constraints to perform image quality degradation processing on multiple third sample images to obtain multiple second sample images includes: labeling each first sample image to determine image quality labeling information corresponding to each first sample image; wherein, the image quality labeling information includes the problem type corresponding to the image quality problem in the first sample image; determining first image distribution information of multiple first sample images under a preset degradation type according to the image quality problem and the problem type, and determining second image distribution information corresponding to the target sample image according to the preset degradation type and the first image distribution information; and using multiple degradation processing algorithms and preset algorithm constraints to perform image quality degradation processing on the third sample images according to the second image distribution information to obtain multiple second sample images.

[0170] According to one or more embodiments of this disclosure, Example 8 provides the method of Example 5, which further includes: optionally, the algorithm constraints include at least constraints on the applicability relationship between the degradation processing algorithm and the object processed by the algorithm and / or constraints on the superposition of multiple degradation processing algorithms.

[0171] According to one or more embodiments of this disclosure, Example 9 provides the method of Example 1, which further includes: optionally, if the image to be detected has an image quality problem, the target feedback information further includes an image quality feedback image; the image quality feedback image identifies a target image region where the image quality problem exists.

[0172] According to one or more embodiments of this disclosure, [Example 10] provides an image quality detection apparatus, comprising: an image detection request module, configured to, in response to an image detection request, acquire a target image to be detected corresponding to the image detection request and target query information corresponding to the target image; wherein the target query information includes information for querying image quality; and an image detection feedback module, configured to process the target image to be detected and the target query information through an image detection model to obtain target feedback information; wherein the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the target image corresponding to the target query information.

[0173] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0174] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0175] The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: in response to an image detection request, acquire a target image to be detected corresponding to the image detection request and target query information corresponding to the target image; wherein the target query information includes information for querying image quality; process the target image to be detected and the target query information through an image detection model to obtain target feedback information; wherein the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the target image corresponding to the target query information.

[0176] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0177] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0178] The units described in the embodiments of this disclosure can be implemented in software or in hardware. The names of the units are not necessarily limiting in certain circumstances; for example, the image detection request module can also be described as "a module for acquiring the image to be detected and target query information".

[0179] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0180] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0181] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0182] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0183] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. An image quality detection method, characterized in that, include: In response to an image detection request, the system acquires the image to be detected corresponding to the image detection request and the target query information corresponding to the image to be detected; wherein, the target query information includes information for querying image quality. The image to be detected and the target query information are processed by an image detection model to obtain target feedback information; wherein, the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the image to be detected corresponding to the target query information.

2. The image quality detection method according to claim 1, characterized in that, The image detection model is trained in the following manner: Acquire multiple target sample images; The deep learning model is trained based on the target sample image and the first image quality description information corresponding to the target sample image to obtain a pre-trained model; The pre-trained model is trained based on the target sample image and the second image quality description information corresponding to the target sample image to obtain an image detection model; wherein, the second image quality description information provides a higher level of detail in describing the image quality than the first image quality description information.

3. The image quality detection method according to claim 2, characterized in that, The step of training the deep learning model based on the target sample image and the first image quality description information to obtain a pre-trained model includes: First image quality description information corresponding to each of the target sample images is determined respectively. An image detection task is constructed based on the target sample images and the first image quality description information. A visual dialogue task is obtained. The visual dialogue task includes a dialogue task associated with the image content and / or the associated scene of the image content. The deep learning model is trained through the image detection task and the visual dialogue task to obtain a pre-trained model.

4. The image quality detection method according to claim 2, characterized in that, The step of training the pre-trained model based on the target sample image and the second image quality description information to obtain an image detection model includes: For a single target sample image, determine the image quality annotation information corresponding to the target sample image; The image quality annotation information is converted into second image quality description information, and the second image quality description information is converted into third image quality description information in various question-and-answer formats; The pre-trained model is trained based on multiple target sample images and the third image quality description information to obtain an image detection model.

5. The image quality detection method according to claim 2, characterized in that, The target sample image includes a first sample image and a second sample image; The acquisition of multiple target sample images includes: Acquire multiple first sample images, and determine multiple third sample images based on at least a portion of the first sample images; wherein the image quality of the third sample images is not lower than that of the first sample images; and, Multiple image degradation algorithms and preset algorithm constraints are used to perform image quality degradation processing on multiple third sample images to obtain multiple second sample images.

6. The image quality detection method according to claim 5, characterized in that, The acquisition of multiple first sample images includes: Multiple candidate images are acquired, and for each candidate image, the image feature values ​​of the candidate image under multiple feature extraction dimensions are determined. For each of the feature extraction dimensions, multiple candidate images are sampled based on the image feature values ​​of multiple candidate images under the feature extraction dimension to obtain multiple first sample images.

7. The image quality detection method according to claim 5, characterized in that, The process involves employing multiple degradation processing algorithms and preset algorithm constraints to perform image quality degradation processing on multiple third sample images to obtain multiple second sample images, including: Each of the first sample images is labeled to determine the image quality labeling information corresponding to each first sample image; wherein, the image quality labeling information includes the problem type corresponding to the image quality problem in the first sample image; Based on the image quality problem and the problem type, determine the first image distribution information of multiple first sample images under a preset degradation type, and determine the second image distribution information corresponding to the target sample image based on the preset degradation type and the first image distribution information; Based on the second image distribution information, various degradation processing algorithms and preset algorithm constraints are used to perform image quality degradation processing on the third sample image to obtain multiple second sample images.

8. The image quality detection method according to claim 5, characterized in that, The algorithm constraints include at least constraints on the applicability of the degradation processing algorithm to the object being processed and / or constraints on the superposition of multiple degradation processing algorithms.

9. The image quality detection method according to claim 1, characterized in that, If the image to be detected has an image quality problem, the target feedback information also includes an image quality feedback image; the image quality feedback image identifies the target image area where the image quality problem exists.

10. An image quality detection device, characterized in that, include: An image detection request module is used to respond to an image detection request by acquiring a target image to be detected corresponding to the image detection request and target query information corresponding to the target image; wherein, the target query information includes information for querying image quality; An image detection feedback module is used to process the image to be detected and the target query information through an image detection model to obtain target feedback information; wherein, the target feedback information includes at least image quality description information; the image quality description information is used to describe the image quality detection result of the image to be detected corresponding to the target query information.

11. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the image quality detection method as described in any one of claims 1-9.

12. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the image quality detection method as described in any one of claims 1-9.

13. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the image quality detection method as described in any one of claims 1-9.

Images