Image quality detection method and device

Abstract

The invention discloses an image quality detection method and device. The method comprises the steps that first attribute information and second attribute information of at least one initial image are acquired, the image content of the initial image comprises a target object needing desensitization processing, the first attribute information is used for representing the definition degree of the corresponding initial image, and the second attribute information is used for at least representing the posture of the corresponding target object; and / or the integrity degree of the target object in the corresponding initial image; in response to the fact that the first attribute information is larger than or equal to a first attribute information threshold value and the second attribute information is larger than or equal to a second attribute information threshold value, a target object in the initial image is subjected to desensitization processing, a target image is obtained, and the image content of the target image comprises the target object subjected to desensitization processing; and detecting the desensitization processing quality of the target image to obtain a detection result. According to the invention, the technical problem of low accuracy of quality detection of image desensitization processing is solved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image quality detection method and device. Background Art

[0002] Currently, image desensitization refers to the process of removing or blurring sensitive information contained in an image while ensuring image usability. This process can avoid leaking information or other sensitive data during the use, storage, or transmission of the image. Therefore, quality testing of image desensitization is very important for determining whether the image meets the desensitization quality indicators.

[0003] In related technologies, image quality detection and whether the desensitization quality indicators are met usually rely on manual judgment. The above method is not only inefficient, but also easily affected by personal subjective judgment, resulting in detection results that are not objective and accurate enough. Therefore, there is still a technical problem of low accuracy in quality detection of image desensitization.

[0004] Currently, no effective solution has been proposed to the above-mentioned technical problem of low accuracy in quality detection of image desensitization processing. Summary of the Invention

[0005] Embodiments of the present invention provide a method and apparatus for detecting image quality, so as to at least solve the technical problem of low accuracy in quality detection of image desensitization processing.

[0006] According to one aspect of an embodiment of the present invention, a method for detecting image quality is provided. The method may include: obtaining first attribute information and second attribute information of at least one initial image, wherein the image content of the initial image includes a target object that needs to be desensitized, the first attribute information is used to indicate the clarity of the corresponding initial image, and the second attribute information is used to at least indicate the posture of the corresponding target object, and / or the completeness of the target object in the corresponding initial image; in response to the first attribute information being greater than or equal to the first attribute information threshold, and the second attribute information being greater than or equal to the second attribute information threshold, performing desensitization on the target object in the initial image to obtain a target image, wherein the image content of the target image includes the target object after desensitization, and the clarity of the target object after desensitization is lower than the clarity of the target object before desensitization; detecting the desensitization quality of the target image to obtain a detection result, wherein the detection result is used to indicate whether the desensitization quality meets the desensitization quality index corresponding to the target image.

[0007] Optionally, the method is applied to an image quality detection model, the quality detection model includes a desensitization processing quality detection module, which detects the desensitization processing quality of the target image to obtain a detection result, including: using the desensitization processing quality detection module to detect the desensitization processing quality of the target image to obtain a desensitization detection rate of the target image, wherein the desensitization detection rate is used to represent the proportion of the target image to the initial image; in response to the desensitization detection rate being greater than a desensitization detection rate threshold, determining that the detection result is that the desensitization processing quality does not meet the desensitization processing quality index; in response to the desensitization detection rate being less than or equal to the desensitization detection rate threshold, determining that the detection result is that the desensitization processing quality meets the desensitization processing quality index.

[0008] Optionally, the quality detection model includes an image discrimination module and an image selection module, and the method further includes: using the image discrimination module to screen out an initial image set from the initial image based on the first attribute information, wherein the initial image set includes a first initial image, and the first initial image is an initial image whose first attribute information is greater than or equal to the first attribute information threshold; using the image selection module to screen out a target image set from the initial image set based on the second attribute information, wherein the target image set includes a target image, and the target image is the first initial image whose second attribute information is greater than or equal to the second attribute information threshold.

[0009] Optionally, an image discrimination module is used to screen out an initial image set from the initial image based on the first attribute information, including: in response to the presence of a second initial image in the initial image whose first attribute information is less than the first attribute information threshold, the image discrimination module is used to delete the second initial image from the initial image to obtain the initial image set.

[0010] Optionally, the image selection module is used to filter out the target image set from the initial image set based on the second attribute information, including: using the image selection module to delete the first initial image whose second attribute information is less than the second attribute information threshold from the initial image set to obtain the target image set.

[0011] Optionally, the desensitization processing quality detection module is used to detect the desensitization processing quality of the target image to obtain the desensitization detection rate of the target image, including: using the desensitization processing quality detection module to obtain the number of initial images from the image discrimination module of the quality detection model, and using the desensitization processing quality detection module to obtain the number of target images from the image selection module of the quality detection model; using the desensitization processing quality detection module to take the ratio of the number of target images to the number of initial images as the desensitization detection rate.

[0012] Optionally, obtaining first attribute information and second attribute information of at least one initial image includes: obtaining the part to be desensitized of the target object in response to detecting a part acquisition instruction; determining the first attribute information and the second attribute information based on the part information of the part to be desensitized, wherein the part information is used to indicate the sensitivity of the part to be desensitized, and the sensitivity of the part to be desensitized is positively correlated with the degree of blur of the desensitization treatment of the part to be desensitized.

[0013] Optionally, the method also includes at least one of the following: determining a first attribute information threshold based on the part information and the first attribute information, wherein the first attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; determining a second attribute information threshold based on the part information and the second attribute information, wherein the second attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; determining a desensitization detection rate threshold corresponding to the desensitization detection rate based on the part information, wherein the size of the desensitization detection rate threshold is positively correlated with the sensitivity of the part to be desensitized.

[0014] Optionally, the method further includes: determining a loss function of the initial quality detection model during training of the initial quality detection model using the initial image samples and the detection result samples; in response to the loss function being greater than a loss function threshold, repeatedly performing the following steps until the loss function is less than or equal to the loss function threshold, and determining the initial quality detection model as the quality detection model: training the initial quality detection model using the initial image samples and the detection result samples, and determining the loss function.

[0015] According to another aspect of an embodiment of the present invention, an image quality detection device is also provided. The device may include: an acquisition unit for acquiring first attribute information and second attribute information of at least one initial image, wherein the image content of the initial image includes a target object that needs to be desensitized, the first attribute information is used to indicate the clarity of the corresponding initial image, and the second attribute information is used to at least indicate the posture of the target object and / or the completeness of the target object in the corresponding initial image; a determination unit for performing desensitization on the target object in the initial image in response to the first attribute information being greater than or equal to the first attribute information threshold and the second attribute information being greater than or equal to the second attribute information threshold, to obtain a target image, wherein the image content of the target image includes the target object after desensitization, and the clarity of the target object after desensitization is lower than the clarity of the target object before desensitization; a detection unit for detecting the desensitization quality of the target image to obtain a detection result, wherein the detection result is used to indicate whether the desensitization quality meets the desensitization quality index corresponding to the target image.

[0016] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the image quality detection method of an embodiment of the present invention.

[0017] According to another aspect of an embodiment of the present invention, a processor is provided, which is configured to run a program, wherein the image quality detection method according to an embodiment of the present invention is executed when the program is run.

[0018] According to another aspect of an embodiment of the present invention, a vehicle is provided, which is used to perform the image quality detection method according to an embodiment of the present invention.

[0019] In an embodiment of the present invention, by obtaining the first attribute information of the initial image and the second attribute information of the target object in the initial image, it is possible to at least determine the clarity of the initial image and the posture of the target object and the degree of completeness in the initial image, and accurately identify which initial images and target objects are suitable for desensitization processing, thereby avoiding indiscriminate desensitization processing. By setting the first attribute information threshold and the second attribute information threshold, it is possible to determine whether the clarity of the initial image and the characteristics of the target object meet the conditions for desensitization processing; when the first attribute information is greater than or equal to the first attribute information threshold, and the second attribute information is greater than or equal to the second attribute information threshold, the target object in the initial image is automatically triggered to perform desensitization processing, reduce the clarity of the target object in the initial image, obtain the target image, perform quality inspection on the desensitized target image, and determine whether the desensitization processing quality index is met. Through the above method, the low efficiency of manual judgment is overcome, and it is easily affected by personal subjective judgment, resulting in inaccurate detection results, and the technical problem of low accuracy of quality detection of image desensitization processing is solved, achieving the technical effect of improving the accuracy of quality detection of image desensitization processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0021] Figure 1 is a flow chart of a method for detecting image quality according to an embodiment of the present invention;

[0022] Figure 2 is a structural diagram of a convolutional neural network model according to an embodiment of the present invention;

[0023] Figure 3is a flowchart of an automated image desensitization testing method based on a convolutional neural network according to an embodiment of the present invention;

[0024] Figure 4 FIG. 4 is a schematic diagram of an image quality detection device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first," "second," and the like in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the numbers used in this way are interchangeable where appropriate so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0027] According to an embodiment of the present invention, an embodiment of a method for detecting image quality is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] Figure 1 FIG. 1 is a flow chart of a method for detecting image quality according to an embodiment of the present invention. Figure 1 As shown, the method may include the following steps:

[0029] Step S102: Acquire first attribute information and second attribute information of at least one initial image.

[0030] In the technical solution provided in the above step S102 of the present invention, the image content of the initial image includes the target object that needs to be desensitized, the first attribute information is used to represent the clarity of the corresponding initial image, and the second attribute information is used to at least represent the posture of the corresponding target object, and / or the completeness of the target object in the corresponding initial image. The target object can be a face, a license plate number, identity information, etc. The first attribute information can be resolution, color channel intensity value. The second attribute information can be facial posture, facial integrity, facial clarity. The above target objects are only for illustration and do not impose specific restrictions on the specific content of the target objects. As long as the target object can be used to represent privacy or sensitive information, it is within the protection scope of the embodiments of the present invention.

[0031] In this embodiment, obtaining the first attribute information and the second attribute information of at least one initial image can provide basic information for the subsequent determination of whether to perform desensitization on the target object in the initial image. The first attribute information and the second attribute information can be used to preliminarily screen which initial images need to be desensitized. For example, if the resolution of an initial image is too low so that the target object (e.g., face) cannot be recognized, then the initial image does not need to be desensitized. On the contrary, if the face in the initial image is clear, the facial posture is normal and is in a recognizable state, the initial image needs to be desensitized to protect personal privacy.

[0032] Optionally, resolution is one of the important indicators for measuring the clarity of the initial image, which involves the density of pixels in the initial image. High resolution means denser pixels, richer details in the initial image, and higher clarity. By obtaining the resolution of at least one initial image, it is possible to determine whether the initial image is clear enough for effective facial recognition and subsequent desensitization processing. The color channel intensity value involves the intensity value distribution of the three color channels of red, green, and blue in the initial image, and can be used to evaluate the color richness and contrast of the initial image, which is necessary to ensure that the initial image is clear enough and that the desensitization processing does not cause serious color distortion.

[0033] Optionally, facial pose can include horizontal rotation, pitch, and tilt angles. Obtaining facial pose can determine the orientation and angle of the face. Facial integrity, which includes the visibility and geometric distortion of parts such as eyebrows, eyes, nose, mouth, and cheek skin, is crucial for the accuracy of facial recognition and the thoroughness of desensitization processing. Facial clarity can be measured using metrics such as Gaussian blur, motion blur, and Laplace variance, and can be used to assess the clarity of the initial image containing the face.

[0034] Optionally, the first attribute information of the initial image can be obtained through an image processing library, and the second attribute information can be used to predict the key point positions through a deep learning model to determine the horizontal rotation angle, pitch angle and tilt angle of the face.

[0035] It should be noted that the content and acquisition method of the above-mentioned first attribute information and second attribute information are only examples and are not specifically limited here. As long as the method can be used to obtain the relevant attribute information of the initial image and the target object, it is within the protection scope of the embodiments of the present invention.

[0036] Step S104 , in response to the first attribute information being greater than or equal to the first attribute information threshold, and the second attribute information being greater than or equal to the second attribute information threshold, performing desensitization processing on the target object in the initial image to obtain a target image.

[0037] In the technical solution provided in the above step S104 of the present invention, the image content of the target image includes the target object after desensitization processing, and the clarity of the target object after desensitization processing is lower than the clarity of the target object before desensitization processing.

[0038] In this embodiment, after obtaining the first attribute information and the second attribute information of at least one initial image, desensitization processing of the target object in the initial image can be automatically triggered if the first attribute information is greater than or equal to a first attribute information threshold, and the second attribute information is greater than or equal to a second attribute information threshold, to obtain a target image. The first attribute information threshold and the second attribute information threshold can be set as needed.

[0039] For example, with respect to resolution, when the initial image's long side is greater than 960 pixels, the minimum detectable face side length is greater than or equal to the image's long side / 60 pixels; when the initial image's short side is greater than 544 pixels, the minimum detectable face side length is greater than 16 pixels; and when the initial image's short side is less than 544 pixels, the minimum detectable face side length is greater than 10 pixels. Therefore, when detecting faces in a 1080P video (1920*1080), according to the above priority, if the initial image's long side is greater than 960 pixels, the minimum detectable face is 1920 / 60 = 32×32 pixels.

[0040] Optionally, for facial posture, it can be determined whether the posture meets the horizontal rotation angle: -45° to 45°; whether the posture meets the pitch angle: -30° to 30°; and whether the posture meets the tilt angle: -45° to 45°.

[0041] Optionally, for completeness, it can be determined whether the geometric distortion is ≤15%; whether the eyebrow visibility is ≥75%; whether the eye visibility is 100%; whether the nose visibility is ≥85%; whether the mouth visibility is 100%; and whether the cheek skin visibility is ≥75%.

[0042] Optionally, for clarity, it may be determined whether Gaussian blur is less than 0.30, whether motion blur is less than 0.26, and whether Laplace variance is ≥ 200.

[0043] Optionally, for each of the three color channels of red, green and blue of the image, the color channel intensity values ​​between 10 and 250 account for no less than 95%.

[0044] It should be noted that the threshold settings corresponding to the above-mentioned resolution, facial posture, completeness, clarity, and color channel intensity values ​​are only for illustration and are not specifically limited here.

[0045] In this embodiment, after the desensitization process is completed, a target image can be obtained. In the target image, the clarity of the target object after the desensitization process is lower than that before the desensitization process, while other parts of the original image are kept unchanged as much as possible to maintain the non-sensitive information and overall structure of the original image.

[0046] Step S106: Detect the desensitization processing quality of the target image to obtain a detection result.

[0047] In the technical solution of step S106 of the present invention, the detection result is used to indicate whether the desensitization quality meets the desensitization quality index corresponding to the target image. The desensitization quality index can be adjusted according to different application scenarios and requirements. For example, it can be facial unrecognizable to ensure that the target image after desensitization cannot be clearly identified, that is, a certain degree of blur or feature deformation is reached, so that the facial features are not obvious. The clarity of the target image can be maintained: although the target object (such as the face) needs to be desensitized, other non-sensitive parts of the target image should maintain their original clarity as much as possible to avoid excessive blurring and loss of target image information.

[0048] It should be noted that the above-mentioned desensitization processing quality indicators are only for illustration and are not specifically limited here. As long as the desensitization processing quality indicators can be used to judge the quality of desensitization processing, they are within the protection scope of the embodiments of the present invention and will not be illustrated one by one here.

[0049] In this embodiment, after desensitizing the target object in the initial image to obtain the target image, the desensitization quality of the target image can be tested, and the first attribute information such as the resolution of the target image and the proportion of color channel intensity values ​​can be recalculated. The difference before and after desensitization is compared to ensure that the clarity of the target object is reduced while the clarity of other non-sensitive areas remains within an acceptable range. Second attribute information, such as facial posture, completeness, and clarity, can be used to check whether these features of the target object after desensitization still meet the second attribute information threshold requirements set before desensitization, while avoiding the recognition of sensitive information.

[0050] Optionally, the desensitization quality of the target image can be tested using a preset algorithm and model to obtain a quantitative desensitization quality score, which can be compared with the desensitization quality index to determine whether the desensitization has achieved the desired effect. If the test result shows that the desensitization quality does not meet the preset desensitization quality index, automatic feedback can be provided and corresponding adjustments can be made, such as reprocessing or optimizing the parameters of the algorithm or model until the specified desensitization quality standard is met.

[0051] In the above steps S102 to S106 of the present application, by obtaining the first attribute information of the initial image and the second attribute information of the target object in the initial image, it is possible to at least determine the clarity of the initial image and the posture of the target object and the degree of completeness in the initial image, and accurately identify which initial images and target objects are suitable for desensitization processing, thereby avoiding indiscriminate desensitization processing. By setting the first attribute information threshold and the second attribute information threshold, it is possible to determine whether the clarity of the initial image and the characteristics of the target object meet the conditions for desensitization processing; when the first attribute information is greater than or equal to the first attribute information threshold, and the second attribute information is greater than or equal to the second attribute information threshold, the target object in the initial image is automatically triggered to perform desensitization processing, the clarity of the target object in the initial image is reduced, the target image is obtained, and the quality of the desensitized target image is tested to determine whether the desensitization processing quality index is met. Through the above method, the low efficiency of manual judgment is overcome, and it is easily affected by personal subjective judgment, resulting in inaccurate detection results, and the technical problem of low accuracy of quality detection of image desensitization processing is solved, and the technical effect of improving the accuracy of quality detection of image desensitization processing is achieved.

[0052] The above method of this embodiment is further introduced below.

[0053] As an optional embodiment, the method is applied to an image quality detection model, and the quality detection model includes a desensitization processing quality detection module. In step S106, the desensitization processing quality of the target image is detected to obtain a detection result, including: using the desensitization processing quality detection module to detect the desensitization processing quality of the target image to obtain a desensitization detection rate of the target image, wherein the desensitization detection rate is used to represent the proportion of the target image to the initial image; in response to the desensitization detection rate being greater than the desensitization detection rate threshold, determining that the detection result is that the desensitization processing quality does not meet the desensitization processing quality index; in response to the desensitization detection rate being less than or equal to the desensitization detection rate threshold, determining that the detection result is that the desensitization processing quality meets the desensitization processing quality index.

[0054] In this embodiment, in the process of detecting the desensitization processing quality of the target image and obtaining the detection result, the desensitization processing quality detection module can be used to detect the desensitization processing quality of the target image and obtain the desensitization detection rate of the target image. When the desensitization detection rate is greater than the desensitization detection rate threshold, it can be determined that the detection result is that the desensitization processing quality does not meet the desensitization processing quality index; when the desensitization detection rate is less than or equal to the desensitization detection rate threshold, it can be determined that the detection result is that the desensitization processing quality meets the desensitization processing quality index. The quality detection model can be a convolutional neural network model. The desensitization processing quality detection module can also be called a desensitization performance calculation module. The desensitization detection rate can be used to indicate the proportion of faces or sensitive information that can still be detected and identified in the target image after desensitization processing, reflecting the effectiveness of the desensitization processing, that is, the amount of sensitive information that can still be detected after processing.

[0055] Optionally, the desensitized image (target image) is input into a desensitization quality detection module, which can run a pre-trained deep learning model or other image analysis algorithm, such as a convolutional neural network model, to analyze the target image, identify the face or other sensitive information therein, and calculate the desensitization detection rate. If the desensitization detection rate is greater than the desensitization detection rate threshold, it indicates that the desensitization effect is not good, and there is still a lot of sensitive information that is not effectively protected. In this case, it is determined that the test result is that the desensitization quality does not meet the desensitization quality index. If the desensitization detection rate is less than or equal to the desensitization detection rate threshold, it means that the desensitization has achieved the expected effect and the sensitive information has been effectively hidden. In this case, it is determined that the test result is that the desensitization quality meets the desensitization quality index.

[0056] Optionally, the convolutional neural network model can consist of a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a fully connected layer, and an output layer. The first convolutional layer can use six convolution kernels, each of which can be 5x5 in size. The output of this layer is six feature maps, each corresponding to a feature extracted by a convolution kernel. The first pooling layer can use max pooling with a window size of 2x2. This layer reduces the dimensionality of the data by downsampling, and the output is six feature maps. The second convolutional layer can use 16 convolution kernels, each of which can be 5x5 in size, and the output of this layer is 16 feature maps. The second pooling layer can also use max pooling with a window size of 2x2. This layer reduces the dimensionality of the data by downsampling, and the output is six feature maps. To integrate local features into global features, the fully connected layer can convert the two-dimensional feature maps output by each of the above layers into one-dimensional vectors, and the number of neurons is 120. The output layer can be a fully connected layer with 14 neurons, each corresponding to a digit 0 or 1. Among them, the output of each layer is the input of the next layer, and the output of each layer is determined by the weight of the current layer and the input of the current layer.

[0057] As an optional embodiment, the quality detection model includes an image discrimination module and an image selection module. The method also includes: using the image discrimination module to screen out an initial image set from the initial image based on the first attribute information, wherein the initial image set includes a first initial image, and the first initial image is an initial image whose first attribute information is greater than or equal to the first attribute information threshold; using the image selection module to screen out a target image set from the initial image set based on the second attribute information, wherein the target image set includes a target image, and the target image is the first initial image whose second attribute information is greater than or equal to the second attribute information threshold.

[0058] In this embodiment, in order to enhance the accuracy of the desensitization process, the image recognition module and the image selection module can be used to perform initial image screening based on the first attribute information and the second attribute information, respectively, to ensure that the initial images that meet specific conditions will be sent to the subsequent desensitization process and quality inspection process.

[0059] Optionally, based on the preliminary screening of the first attribute information, an image discrimination module can be used to screen an initial image set from the initial images based on the first attribute information. Specifically, the initial images whose first attribute information is greater than or equal to the first attribute information threshold can be screened. This screening by the image discrimination module reduces the waste of computational resources involved in unnecessary processing of low-quality initial images or those that do not require desensitization, thereby improving the efficiency of the entire desensitization process.

[0060] Optionally, deep filtering based on the second attribute information can be performed using an image selection module. Based on the second attribute information, a target image set can be selected from the initial image set. Specifically, the target image set is selected, specifically, the first initial image whose second attribute information is greater than or equal to the second attribute threshold. This filtering by the image selection module ensures the accuracy of the desensitization process and avoids processing initial images that do not contain sensitive information or whose sensitive information is blurred.

[0061] Optionally, if the first attribute information is greater than or equal to the first attribute information threshold, a value (label) of 1 can be assigned, otherwise a value of 0 can be assigned. In other words, if the first attribute information is greater than or equal to the first attribute information threshold, a value of 1 can be assigned, otherwise a value of 0 can be assigned. Similarly, if the second attribute information is greater than or equal to the second attribute information threshold, a value of 1 can be assigned, otherwise a value of 0 can be assigned.

[0062] As an optional embodiment, an image discrimination module is used to screen out an initial image set from the initial image based on the first attribute information, including: in response to the presence of a second initial image in the initial image whose first attribute information is less than the first attribute information threshold, the image discrimination module is used to delete the second initial image from the initial image to obtain the initial image set.

[0063] In this embodiment, in the process of using the image discrimination module to screen out the initial image set from the initial image based on the first attribute information, if there is a second initial image in the initial image whose first attribute information is less than the first attribute information threshold, the image discrimination module can be used to delete the second initial image from the initial image to obtain the initial image set.

[0064] Optionally, first attribute information is extracted from the initial image, including but not limited to resolution, color channel intensity value ratio, etc., and the extracted first attribute information can be judged according to a preset first attribute information threshold. If the first attribute information is lower than the first attribute information threshold, that is, the quality of the initial image is too low or does not meet the desensitization processing requirements, it can be marked as a second initial image. The image discrimination module is used to delete the second initial image, which can ensure the targeted desensitization processing, reduce the possibility of processing invalid or low-quality initial images, and improve the stability and reliability of the desensitization processing.

[0065] As an optional implementation method, an image selection module is used to screen out a target image set from the initial image set based on the second attribute information, including: using the image selection module to delete the first initial image whose second attribute information is less than the second attribute information threshold from the initial image set to obtain the target image set.

[0066] In this embodiment, in the process of using the image selection module to filter out the target image set from the initial image set based on the second attribute information, the image selection module can be used to delete the first initial image whose second attribute information is less than the second attribute information threshold from the initial image set to obtain the target image set.

[0067] Optionally, initial images are loaded from the initial image set output by the image discrimination module. These initial images have passed the preliminary screening based on the first attribute information, that is, the clarity and quality have reached a certain standard. Each first initial image is deeply analyzed to obtain the second attribute information of the target object, including but not limited to features such as facial posture, completeness, and clarity. By comparing the second attribute information of each first initial image with the second attribute information threshold, images whose target object features (second attribute information) are lower than the set threshold are identified. These images may contain blurred faces, obscured targets, or parts that do not contain sensitive information at all, so desensitization is not required. The image selection module can automatically delete the above-mentioned first initial images from the initial image set. After the above-mentioned screening and deletion process, the remaining images will form a target image set. The target object features (second attribute information) in these images meet or exceed the second attribute information threshold, that is, they are considered to be clear and complete enough and need to be desensitized to ensure the security and privacy protection of sensitive information.

[0068] As an optional implementation method, a desensitization processing quality detection module is used to detect the desensitization processing quality of the target image to obtain the desensitization detection rate of the target image, including: using the desensitization processing quality detection module to obtain the number of initial images from the image discrimination module of the quality detection model, and using the desensitization processing quality detection module to obtain the number of target images from the image selection module of the quality detection model; using the desensitization processing quality detection module to take the ratio of the number of target images to the number of initial images as the desensitization detection rate.

[0069] In this embodiment, in the process of using the desensitization processing quality detection module to detect the desensitization processing quality of the target image and obtain the desensitization detection rate of the target image, the desensitization processing quality detection module can be used to obtain the number of initial images from the image discrimination module of the quality detection model, and the desensitization processing quality detection module can be used to obtain the number of target images from the image selection module of the quality detection model; the desensitization processing quality detection module can be used to take the ratio of the number of target images to the number of initial images as the desensitization detection rate.

[0070] Alternatively, the desensitization detection rate, also referred to as the face detection rate, can be determined by the following formula:

[0071] Face detection rate = M_1 / M_0×100%

[0072] Among them, M_0 can be used to represent the number of initial images, and M_1 can be used to represent the number of target images after desensitization.

[0073] Optionally, a comparison of the desensitization detection rate with a preset desensitization detection rate threshold can be used to determine whether the desensitization process meets established privacy protection standards. If the desensitization detection rate is higher than the desensitization detection rate threshold, error analysis is required to reduce the desensitization detection rate. Conversely, if the desensitization detection rate is lower than or equal to the desensitization detection rate threshold, the desensitization process is effective and sensitive information is effectively hidden.

[0074] As an optional embodiment, step S102, obtaining first attribute information and second attribute information of at least one initial image, includes: in response to detecting a part acquisition instruction, obtaining the part to be desensitized of the target object; based on the part information of the part to be desensitized, determining the first attribute information and the second attribute information, wherein the part information is used to indicate the sensitivity of the part to be desensitized, and the sensitivity of the part to be desensitized is positively correlated with the degree of blur of the desensitization processing of the part to be desensitized.

[0075] In this embodiment, during the process of acquiring the first attribute information and the second attribute information of at least one initial image, the target object's to-be-desensitized portion can be acquired upon detecting a portion acquisition instruction; and the first attribute information and the second attribute information can be determined based on the portion information of the to-be-desensitized portion. The portion information can be used to indicate the sensitivity of the to-be-desensitized portion, and the sensitivity of the to-be-desensitized portion is positively correlated with the degree of blurring during the desensitization process for the to-be-desensitized portion.

[0076] Optionally, according to the part acquisition instruction, image analysis technology (such as deep learning, image recognition algorithm, etc.) can be used to identify and mark the parts of the target object to be desensitized in the initial image, such as facial posture, eyebrows, eyes, etc. For the marked parts to be desensitized, first attribute information and second attribute information can be determined to ensure that the specific needs and characteristics of the parts to be desensitized are fully considered.

[0077] Optionally, when determining the first attribute information and the second attribute information, the sensitivity of the area to be desensitized can be taken into account. The higher the sensitivity, the higher the degree of blur required during desensitization to ensure the unrecognizable nature of the sensitive information. This positive correlation ensures that the desensitization process can meet the needs of privacy protection while avoiding unnecessary excessive blurring of the original image, thereby maintaining the usability and information integrity of the original image.

[0078] As an optional embodiment, the method also includes at least one of the following: determining a first attribute information threshold based on the part information and the first attribute information, wherein the first attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; determining a second attribute information threshold based on the part information and the second attribute information, wherein the second attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; determining a desensitization detection rate threshold corresponding to the desensitization detection rate based on the part information, wherein the size of the desensitization detection rate threshold is positively correlated with the sensitivity of the part to be desensitized.

[0079] In this embodiment, in order to conduct targeted analysis of different sensitive areas and ensure that the desensitization process is both effective and not excessive, thereby finding a balance between protecting privacy and retaining available target images, the first attribute information threshold, the second attribute information threshold, and the desensitization detection rate threshold corresponding to the desensitization detection rate can be dynamically adjusted according to the sensitivity of the area to be desensitized.

[0080] Optionally, the first attribute information threshold can be dynamically adjusted based on the area information and sensitivity. Highly sensitive areas (such as the face) will have a lower first attribute information threshold, meaning that images with clear areas of these areas can be screened out for desensitization, even if the overall image quality is slightly lower.

[0081] Optionally, the second attribute information threshold can be dynamically adjusted based on the location information and sensitivity. For highly sensitive areas, a lower second attribute information threshold can be set to ensure that even if the integrity or posture of a highly sensitive area in the image is slightly poor, it will be selected for desensitization, minimizing the risk of sensitive information leakage.

[0082] Optionally, the setting of the desensitization detection rate threshold is directly related to the strictness of the desensitization process and the evaluation of its effectiveness. The size of the desensitization detection rate threshold is positively correlated with the sensitivity of the area to be desensitized. That is, for more sensitive areas, a higher desensitization detection rate threshold is set. This means that even after desensitization, a certain proportion of sensitive areas are allowed to be detected, but this proportion will be dynamically adjusted based on the sensitivity of the area to ensure that the overall privacy protection level meets the requirements.

[0083] As an optional embodiment, the method also includes: determining the loss function of the initial quality detection model during the process of training the initial quality detection model using the initial image samples and the detection result samples; in response to the loss function being greater than the loss function threshold, repeatedly performing the following steps until the loss function is less than or equal to the loss function threshold, and determining the initial quality detection model as the quality detection model: training the initial quality detection model using the initial image samples and the detection result samples, and determining the loss function.

[0084] In this embodiment, the loss function of the initial quality detection model can be determined during the process of training the initial quality detection model using the initial image samples and the detection result samples. Based on the expectations for the performance of the initial quality detection model, a loss function threshold can be pre-set, that is, the maximum error range that the initial quality detection model can accept after training is completed, to determine whether the initial quality detection model has achieved the required training effect.

[0085] Optionally, after each round of initial quality detection model training is completed, the loss function of the current initial quality detection model is calculated. If the loss function is greater than the loss function threshold, it indicates that the performance of the initial quality detection model still does not meet the requirements and needs to be further optimized. At this point, the initial quality detection model training process can be repeated, that is, the initial image samples and detection result samples are reused for training in order to obtain a better initial quality detection model parameter configuration, until the loss function of the initial quality detection model drops below the loss function threshold. At this point, it means that the performance of the initial quality detection model has met the established accuracy requirements. The training process can be stopped, and the current initial quality detection model can be determined as the final quality detection model for subsequent actual target image desensitization processing and quality detection tasks.

[0086] Alternatively, the loss function can be determined by the following formula:

[0087]

[0088] Among them, M can be used to represent the number of initial image samples; y SL (m) can be used to represent the expected output of the initial image label; d SL (m) can be used to represent the actual output of the initial image label; L(Θ) can be used to represent the mean square error between the actual output and the expected output.

[0089] Optionally, the smaller the loss function is, the better the initial quality detection model parameters are trained, and the closer the actual output of the initial quality detection model is to the expected output.

[0090] In an embodiment of the present invention, by obtaining the first attribute information of the initial image and the second attribute information of the target object in the initial image, it is possible to at least determine the clarity of the initial image and the posture of the target object and the degree of completeness in the initial image, and accurately identify which initial images and target objects are suitable for desensitization processing, thereby avoiding indiscriminate desensitization processing. By setting the first attribute information threshold and the second attribute information threshold, it is possible to determine whether the clarity of the initial image and the characteristics of the target object meet the conditions for desensitization processing; when the first attribute information is greater than or equal to the first attribute information threshold, and the second attribute information is greater than or equal to the second attribute information threshold, the target object in the initial image is automatically triggered to perform desensitization processing, reduce the clarity of the target object in the initial image, obtain the target image, perform quality inspection on the desensitized target image, and determine whether the desensitization processing quality index is met. Through the above method, the low efficiency of manual judgment is overcome, and it is easily affected by personal subjective judgment, resulting in inaccurate detection results, and the technical problem of low accuracy of quality detection of image desensitization processing is solved, achieving the technical effect of improving the accuracy of quality detection of image desensitization processing.

[0091] The technical solutions of the embodiments of the present invention are described below with reference to preferred implementation methods.

[0092] Currently, multi-functional facial images collected by vehicle sentry mode, shadow mode, and other methods require facial desensitization on the vehicle itself before uploading to the cloud. Data desensitization requires certain quality requirements for the input images. If they do not meet these quality requirements, desensitization is not performed and they are not counted as inspected. However, in related technologies, manual screening of inspected images is labor-intensive, inefficient, and subjectively leads to low accuracy. Therefore, the technical issue of low quality testing accuracy after image desensitization still exists.

[0093] The embodiment of the present invention proposes an automated image desensitization test method based on a convolutional neural network, which is applied to an automated image desensitization test system, and includes an image discrimination module, an image selection module, and a desensitization performance calculation module. The input of the image discrimination module is a facial image before desensitization, which is used to screen facial images that meet the desensitization quality requirements. The image selection module can be used to screen images that meet the facial desensitization quality requirements. The desensitization performance calculation module can be used to calculate the facial desensitization detection rate, thereby solving the technical problem of low accuracy of quality detection of image desensitization processing, and achieving the technical effect of improving the accuracy of quality detection of image desensitization processing.

[0094] The following is a further introduction to the embodiments of the present invention.

[0095] Figure 2 is a structural diagram of a convolutional neural network model according to an embodiment of the present invention, such as Figure 2 As shown, it includes: convolution layer 201, pooling layer 202, convolution layer 203, pooling layer 204, fully connected layer 205, and output layer 206.

[0096] Convolution layer 201 (the first convolution layer) can use 6 convolution kernels, each of which is usually 5x5 in size. The output of this layer is 6 feature maps, each of which corresponds to a feature extracted by a convolution kernel.

[0097] The pooling layer 202 (the first pooling layer) can use maximum pooling with a window size of 2x2. This layer reduces the dimension of the data by downsampling and outputs 6 feature maps.

[0098] Convolution layer 203 (the second convolution layer) can use 16 convolution kernels, each of which is usually 5x5 in size. The output of this layer is 16 feature maps.

[0099] The pooling layer 204 (the second pooling layer) can also use maximum pooling with a window size of 2x2. This layer reduces the dimension of the data by downsampling and outputs 6 feature maps.

[0100] The fully connected layer 205 can convert the two-dimensional feature maps output by the previous layers into one-dimensional vectors. This step can integrate local features into global features. The number of neurons is 120.

[0101] The output layer 206 is a fully connected layer with 14 neurons, each neuron corresponds to a number 0 or 1. The output of each layer is the input of the next layer, and the output of each layer is determined by the weights and input of the current layer.

[0102] Alternatively, the loss function can be determined by the following formula:

[0103]

[0104] Among them, M can be used to represent the number of initial image samples; y SL (m) can be used to represent the expected output of the initial image label; d SL (m) can be used to represent the actual output of the initial image label; L(Θ) can be used to represent the mean square error between the actual output and the expected output.

[0105] Alternatively, the smaller the loss function, the better the parameter training is, and the closer the actual output of the convolutional neural network model is to the expected output.

[0106] Figure 3 is a flow chart of an automated image desensitization testing method based on a convolutional neural network according to an embodiment of the present invention. Figure 3 As shown, the following steps are included:

[0107] Step S301: Filter images using an image recognition module.

[0108] In this embodiment, the input of the image discrimination module is the facial image before desensitization, which is used to screen facial images that meet the desensitization quality requirements.

[0109] Step S302: The image selection module judges the results of the neural network image discrimination module to determine whether all the results are 1.

[0110] In this embodiment, the image selection module is used to screen images that meet the quality requirements of facial desensitization. It can determine whether all the results output by the image discrimination module are 1. All 1 means that for each image, the output of the neural network for quality evaluation and target feature recognition is 1, which means "qualified". If any indicator is lower than the set threshold, the output is 0, indicating that the image fails to pass the screening. That is, for images that meet the first attribute information greater than or equal to the first attribute information threshold, a value (label) of 1 can be assigned, otherwise a value of 0 can be assigned. Similarly, for images that meet the second attribute information greater than or equal to the second attribute information threshold, a value of 1 can be assigned, otherwise a value of 0 can be assigned.

[0111] Step S303: input the image into the desensitization performance calculation module.

[0112] In this embodiment, if the image passes the dual screening of the image discrimination module and the image selection module, the next step is to input the image into the desensitization performance calculation module to prepare for desensitization processing.

[0113] Step S304: Calculate the facial desensitization detection rate using the desensitization performance calculation module.

[0114] In this embodiment, the desensitization performance calculation module can calculate the detection rate after facial desensitization to evaluate the effect of the desensitization process.

[0115] Alternatively, the desensitization detection rate, also referred to as the face detection rate, can be determined by the following formula:

[0116] Face detection rate = M_1 / M_0×100%

[0117] Among them, M_0 can be used to represent the number of initial images, and M_1 can be used to represent the number of target images after desensitization.

[0118] In an embodiment of the present invention, after the image discrimination module is trained, it has the ability to screen images that meet the quality requirements of facial desensitization. The image before desensitization is input into the image discrimination module for screening. The image selection module deletes the desensitized image corresponding to the pre-desensitization image that is not all 1 based on the input result of the image discrimination module. The image selection module inputs the desensitized image with the result of all 1 to the desensitization performance calculation module. The desensitization performance calculation module calculates the facial desensitization detection rate. Through the above steps, the degree of automation of the facial desensitization test can be effectively improved, manual operations can be reduced, and test efficiency can be improved. At the same time, the error rate caused by manual intervention can be reduced, and the accuracy of the test can be improved.

[0119] According to an embodiment of the present invention, an image quality detection device is further provided. It should be noted that the image quality detection device can be used to execute the image quality detection method in the embodiment.

[0120] Figure 4 FIG is a schematic diagram of an image quality detection device according to an embodiment of the present invention. Figure 4 As shown, the image quality detection device 400 may include: an acquisition unit 402 , a determination unit 404 and a detection unit 406 .

[0121] An acquisition unit 402 is used to acquire first attribute information and second attribute information of at least one initial image, wherein the image content of the initial image includes a target object that needs to be desensitized, the first attribute information is used to indicate the clarity of the corresponding initial image, and the second attribute information is used to indicate at least the posture of the target object and / or the completeness of the target object in the corresponding initial image.

[0122] Determination unit 404 is used to perform desensitization processing on the target object in the initial image in response to the first attribute information being greater than or equal to the first attribute information threshold and the second attribute information being greater than or equal to the second attribute information threshold, to obtain a target image, wherein the image content of the target image includes the target object after desensitization processing, and the clarity of the target object after desensitization processing is lower than the clarity of the target object before desensitization processing.

[0123] The detection unit 406 is used to detect the desensitization processing quality of the target image and obtain a detection result, wherein the detection result is used to indicate whether the desensitization processing quality meets the desensitization processing quality index corresponding to the target image.

[0124] Optionally, the detection unit 406 includes: a first detection module, used to use the desensitization processing quality detection module to detect the desensitization processing quality of the target image to obtain the desensitization detection rate of the target image, wherein the desensitization detection rate is used to represent the proportion of the target image to the initial image; a first determination module, used to determine that the detection result is that the desensitization processing quality does not meet the desensitization processing quality index in response to the desensitization detection rate being greater than the desensitization detection rate threshold; a second determination module, used to determine that the detection result is that the desensitization processing quality meets the desensitization processing quality index in response to the desensitization detection rate being less than or equal to the desensitization detection rate threshold.

[0125] Optionally, the image quality detection device 400 may further include: a first screening unit, used to use the image discrimination module to screen out an initial image set from the initial image based on the first attribute information, wherein the initial image set includes a first initial image, and the first initial image is an initial image whose first attribute information is greater than or equal to the first attribute information threshold; a second screening unit, used to use the image selection module to screen out a target image set from the initial image set based on the second attribute information, wherein the target image set includes a target image, and the target image is the first initial image whose second attribute information is greater than or equal to the second attribute information threshold.

[0126] Optionally, the first screening unit includes: a first deletion module for, in response to the presence of a second initial image in the initial image whose first attribute information is less than the first attribute information threshold, deleting the second initial image from the initial image using the image discrimination module to obtain an initial image set.

[0127] Optionally, the second screening unit includes: a second deletion module, configured to use the image selection module to delete the first initial image whose second attribute information is less than the second attribute information threshold from the initial image set to obtain the target image set.

[0128] Optionally, the first detection module includes: an acquisition submodule, which is used to use the desensitization processing quality detection module to obtain the number of initial images from the image discrimination module of the quality detection model, and use the desensitization processing quality detection module to obtain the number of target images from the image selection module of the quality detection model; a determination submodule, which is used to use the desensitization processing quality detection module to take the ratio of the number of target images to the number of initial images as the desensitization detection rate.

[0129] Optionally, the acquisition unit 402 includes: a first acquisition module, used to acquire the part to be desensitized of the target object in response to detecting a part acquisition instruction; a third determination module, used to determine the first attribute information and the second attribute information based on the part information of the part to be desensitized, wherein the part information is used to indicate the sensitivity of the part to be desensitized, and the sensitivity of the part to be desensitized is positively correlated with the degree of fuzziness of the desensitization processing of the part to be desensitized.

[0130] Optionally, the image quality detection device 400 may further include: a first determination unit for determining a first attribute information threshold based on the part information and the first attribute information, wherein the first attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; a second determination unit for determining a second attribute information threshold based on the part information and the second attribute information, wherein the second attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; a third determination unit for determining a desensitization detection rate threshold corresponding to the desensitization detection rate based on the part information, wherein the size of the desensitization detection rate threshold is positively correlated with the sensitivity of the part to be desensitized.

[0131] Optionally, the image quality detection device 400 may further include: a fourth determination unit, used to determine the loss function of the initial quality detection model during the process of training the initial quality detection model using the initial image samples and the detection result samples; a fifth determination unit, used to, in response to the loss function being greater than the loss function threshold, repeatedly perform the following steps until the loss function is less than or equal to the loss function threshold, and determine the initial quality detection model as the quality detection model: train the initial quality detection model using the initial image samples and the detection result samples, and determine the loss function.

[0132] In an embodiment of the present invention, first attribute information and second attribute information of at least one initial image are acquired by an acquisition unit 402, wherein the image content of the initial image includes a target object that needs to be desensitized, the first attribute information is used to indicate the clarity of the corresponding initial image, and the second attribute information is used to at least indicate the posture of the target object, and / or the completeness of the target object in the corresponding initial image; in response to the first attribute information being greater than or equal to the first attribute information threshold, and the second attribute information being greater than or equal to the second attribute information threshold, a determination unit 404 performs desensitization on the target object in the initial image to obtain a target image, wherein the image content of the target image includes the target object after desensitization, and the clarity of the target object after desensitization is lower than the clarity of the target object before desensitization; the desensitization quality of the target image is detected by a detection unit 406 to obtain a detection result, wherein the detection result is used to indicate whether the desensitization quality meets the desensitization quality index corresponding to the target image, thereby solving the technical problem of low accuracy of quality detection of image desensitization and achieving the technical effect of improving the accuracy of quality detection of image desensitization.

[0133] According to an embodiment of the present invention, a computer-readable storage medium is further provided, which includes a stored program, wherein the program executes the quality detection method for image desensitization processing in the embodiment.

[0134] According to an embodiment of the present invention, a processor is further provided, which is used to run a program, wherein the quality detection method for image desensitization processing in the embodiment is executed when the program is running.

[0135] According to an embodiment of the present invention, a vehicle is also provided, which is used to execute the quality detection method for image desensitization processing in an embodiment of the present invention.

[0136] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application, such as the data for inspection, are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0137] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0139] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0140] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0141] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM), random access memory (RAM), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.

[0142] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting image quality, characterized in that: include: Acquiring first attribute information and second attribute information of at least one initial image, wherein the image content of the initial image includes a target object requiring desensitization processing, the first attribute information is used to indicate clarity of the corresponding initial image, and the second attribute information is used to indicate at least a posture of the corresponding target object and / or a degree of completeness of the target object in the corresponding initial image; In response to the first attribute information being greater than or equal to a first attribute information threshold, and the second attribute information being greater than or equal to a second attribute information threshold, performing desensitization processing on the target object in the initial image to obtain a target image, wherein image content of the target image includes the target object after the desensitization processing, and a clarity of the target object after the desensitization processing is lower than a clarity of the target object before the desensitization processing; The desensitization processing quality of the target image is detected to obtain a detection result, wherein the detection result is used to indicate whether the desensitization processing quality meets the desensitization processing quality index corresponding to the target image.

2. The method according to claim 1, characterized in that The method is applied to an image quality detection model, wherein the quality detection model includes a desensitization processing quality detection module, which detects the desensitization processing quality of at least one target image and obtains a detection result, including: Using the desensitization processing quality detection module, the desensitization processing quality of the target image is detected to obtain a desensitization detection rate of the target image, wherein the desensitization detection rate is used to represent the ratio of the target image to the initial image; In response to the desensitization detection rate being greater than a desensitization detection rate threshold, determining that the detection result is that the desensitization processing quality does not meet the desensitization processing quality indicator; In response to the desensitization detection rate being less than or equal to the desensitization detection rate threshold, it is determined that the detection result is that the desensitization processing quality meets the desensitization processing quality indicator.

3. The method according to claim 2, characterized in that The quality detection model includes an image discrimination module and an image selection module, and the method further includes: Using the image discrimination module, based on the first attribute information, screen an initial image set from the initial images, wherein the initial image set includes a first initial image, and the first initial image is the initial image for which the first attribute information is greater than or equal to the first attribute information threshold; Using the image selection module, based on the second attribute information, a target image set is screened out from the initial image set, wherein the target image set includes the target image, and the target image is the first initial image whose second attribute information is greater than or equal to the second attribute information threshold.

4. The method according to claim 3, characterized in that Utilizing the image discrimination module and based on the first attribute information, screening an initial image set from the initial images, including: In response to the presence of a second initial image in the initial image whose first attribute information is less than the first attribute information threshold, the image discrimination module is used to delete the second initial image from the initial image to obtain the initial image set.

5. The method according to claim 3, characterized in that Utilizing the image selection module to select a target image set from the initial image set based on the second attribute information includes: The image selection module is used to delete the first initial images whose second attribute information is less than the second attribute information threshold from the initial image set to obtain the target image set.

6. The method according to claim 2, characterized in that The desensitization processing quality detection module is used to detect the desensitization processing quality of the target image to obtain the desensitization detection rate of the target image, including: Using the desensitization processing quality detection module, obtain the number of the initial images from the image discrimination module of the quality detection model, and using the desensitization processing quality detection module, obtain the number of the target images from the image selection module of the quality detection model; The desensitization processing quality detection module is used to take the ratio of the number of the target images to the number of the initial images as the desensitization detection rate.

7. The method according to claim 1, characterized in that Acquiring first attribute information and second attribute information of at least one initial image, including: In response to detecting a part acquisition instruction, acquiring the part to be desensitized of the target object; Based on the part information of the part to be desensitized, the first attribute information and the second attribute information are determined, wherein the part information is used to indicate the sensitivity of the part to be desensitized, and the sensitivity of the part to be desensitized is positively correlated with the fuzziness of the desensitization treatment of the part to be desensitized.

8. The method according to claim 7, characterized in that The method further comprises at least one of the following: Determining a first attribute information threshold based on the part information and the first attribute information, wherein the first attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; Determining a second attribute information threshold based on the part information and the second attribute information, wherein the second attribute information threshold is negatively correlated with the sensitivity of the part to be desensitized; Based on the part information, a desensitization detection rate threshold corresponding to the desensitization detection rate is determined, wherein the size of the desensitization detection rate threshold is positively correlated with the sensitivity of the part to be desensitized.

9. The method according to any one of claims 1 to 8, characterized in that The method further comprises: In the process of training the initial quality detection model using the initial image samples and the detection result samples, determining a loss function of the initial quality detection model; In response to the loss function being greater than the loss function threshold, repeatedly performing the following steps until the loss function is less than or equal to the loss function threshold, and determining the initial quality detection model as the quality detection model: The initial quality detection model is trained using the initial image samples and the detection result samples, and the loss function is determined.

10. A quality detection device for image desensitization processing, characterized in that: include: an acquiring unit, configured to acquire first attribute information and second attribute information of at least one initial image, wherein the image content of the initial image includes a target object requiring desensitization processing, the first attribute information is used to indicate the clarity of the corresponding initial image, and the second attribute information is used to indicate at least the posture of the target object and / or the completeness of the target object in the corresponding initial image; a determining unit, configured to, in response to the first attribute information being greater than or equal to a first attribute information threshold and the second attribute information being greater than or equal to a second attribute information threshold, perform desensitization processing on the target object in the initial image to obtain a target image, wherein image content of the target image includes the target object after the desensitization processing, and a clarity of the target object after the desensitization processing is lower than a clarity of the target object before the desensitization processing; The testing unit is used to detect the desensitization processing quality of the target image and obtain a detection result, wherein the detection result is used to indicate whether the desensitization processing quality meets the desensitization processing quality index corresponding to the target image.

Images