WO2022187991A1 - Procédé de traitement d'image, appareil de traitement d'image, dispositif électronique et support d'enregistrement - Google Patents

Procédé de traitement d'image, appareil de traitement d'image, dispositif électronique et support d'enregistrement Download PDF

Info

Publication number
WO2022187991A1
WO2022187991A1 PCT/CN2021/079485 CN2021079485W WO2022187991A1 WO 2022187991 A1 WO2022187991 A1 WO 2022187991A1 CN 2021079485 W CN2021079485 W CN 2021079485W WO 2022187991 A1 WO2022187991 A1 WO 2022187991A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
similarity
space
training
loss
Prior art date
Application number
PCT/CN2021/079485
Other languages
English (en)
Chinese (zh)
Inventor
唐煜
Original Assignee
深圳市欢太科技有限公司
Oppo广东移动通信有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市欢太科技有限公司, Oppo广东移动通信有限公司 filed Critical 深圳市欢太科技有限公司
Priority to CN202180088292.7A priority Critical patent/CN116964582A/zh
Priority to PCT/CN2021/079485 priority patent/WO2022187991A1/fr
Publication of WO2022187991A1 publication Critical patent/WO2022187991A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules

Definitions

  • the present application relates to the technical field of image processing, and in particular, to an image processing method, an image processing apparatus, an electronic device, and a storage medium.
  • Embodiments of the present application provide an image processing method, an image processing apparatus, an electronic device, and a storage medium.
  • the image processing method of the embodiment of the present application includes: acquiring an image to be processed; performing noise processing on the image to be processed by using a preset noise model to obtain a target image, where the target image and the image to be processed are in a first space
  • the first similarity is greater than the first preset processing similarity
  • the second similarity between the target image and the to-be-processed image in the second space is less than the second preset processing similarity, wherein the first space is the original One of the space and the feature space, and the second space is the other one of the original space and the feature space.
  • the image processing apparatus of the embodiment of the present application includes a first acquisition module and a first noise addition module.
  • the first acquisition module is used to acquire the image to be processed.
  • the first noise adding module is configured to perform noise processing on the image to be processed by using a preset noise model to obtain a target image, and the first similarity between the target image and the image to be processed in the first space is greater than the first similarity
  • the preset processing similarity, the second similarity between the target image and the to-be-processed image in the second space is less than the second preset processing similarity, where the first space is one of the original space and the feature space , and the second space is another one of the original space and the feature space.
  • the electronic device of the embodiment of the present application includes one or more processors and a memory, and the memory stores a computer program.
  • the computer program is executed by the processor, the steps of the image processing method of the above embodiment are implemented.
  • the computer-readable storage medium of the embodiment of the present application stores a computer program.
  • the program is executed by the processor, the steps of the image processing method of the above-described embodiment are realized.
  • the similarity between the target image and the image to be processed in the first space is relatively high and the similarity in the second space is relatively low.
  • the information of the to-be-processed image cannot be obtained from the target image in the original space, but the user can obtain the information of the to-be-processed image from the target image in the original space, or the user cannot obtain the information of the to-be-processed image from the target image in the original space, but In the feature space, the machine can obtain the information of the to-be-processed image according to the target image, thereby reducing the information leakage of the to-be-processed image caused by the target image.
  • FIG. 1 is a schematic flowchart of an image processing method according to an embodiment of the present application.
  • FIG. 2 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • FIG. 3 is a schematic diagram of an electronic device according to an embodiment of the present application.
  • FIG. 4 is a schematic flowchart of an image processing method according to an embodiment of the present application.
  • FIG. 5 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • FIG. 6 is a schematic flowchart of an image processing method according to an embodiment of the present application.
  • FIG. 7 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • FIG. 8 is a schematic flowchart of an image processing method according to an embodiment of the present application.
  • FIG. 9 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • FIG. 10 is a schematic flowchart of an image processing method according to an embodiment of the present application.
  • FIG. 11 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • FIG. 12 is a schematic flowchart of an image processing method according to an embodiment of the present application.
  • FIG. 13 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • 15 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • 16 is a schematic flowchart of an image processing method according to an embodiment of the present application.
  • FIG. 17 is a schematic diagram of an image processing apparatus according to an embodiment of the present application.
  • the image processing apparatus 100 the first acquisition module 10, the first noise addition module 20, the second acquisition module 30, the second noise addition module 40, the third acquisition module 50, the first acquisition unit 51, the first acquisition subunit 511, Fourth acquisition module 60 , second acquisition unit 61 , first determination subunit 611 , second acquisition subunit 612 , cropping subunit 613 , alignment subunit 614 , adjustment module 70 , determination unit 71 , second determination subunit 711 , a third determination subunit 712, an adjustment unit 72, a fifth acquisition module 80, a first selection unit 81, and a second selection unit 82;
  • Electronic device 200 processor 201 , memory 202 .
  • the present application provides an image processing method, the image processing method includes:
  • 020 Perform noise processing on the image to be processed by using a preset noise model to obtain a target image, where the first similarity between the target image and the image to be processed in the first space is greater than the first preset similarity for processing, and the target image and the image to be processed have a first similarity greater than the first preset similarity.
  • the second similarity in the second space is smaller than the second preset processing similarity, where the first space is one of the original space and the feature space, and the second space is the other of the original space and the feature space.
  • the image processing method of the embodiment of the present application can be implemented by the image processing apparatus 100 of the embodiment of the present invention.
  • the image processing apparatus 100 includes a first acquisition module 10 and a first noise addition module 20 .
  • the first acquisition module 10 is used for acquiring the image to be processed.
  • the first noise adding module 20 is configured to perform noise processing on the image to be processed by using a preset noise model to obtain a target image, and the first similarity between the target image and the image to be processed in the first space is greater than the first preset processing similarity,
  • the second similarity between the target image and the image to be processed in the second space is smaller than the second preset processing similarity, where the first space is one of the original space and the feature space, and the second space is the original space and the feature space Another kind of.
  • the image processing method of the embodiment of the present application can also be implemented by the electronic device 200 of the embodiment of the present invention.
  • the electronic device 200 includes one or more processors 201 and a memory 202.
  • the memory 202 stores a computer program.
  • the computer program is executed by the processor 201, the steps of the image processing method in the above-mentioned embodiments are implemented.
  • the similarity between the target image and the image to be processed in the first space is high and the similarity in the second space is low.
  • the machine can be made in the feature space.
  • the information of the to-be-processed image cannot be obtained from the target image, but the user can obtain the information of the to-be-processed image from the target image in the original space, or the user cannot obtain the information of the to-be-processed image from the target image in the original space, but the machine In the feature space, the information of the to-be-processed image can be obtained according to the target image, thereby reducing the information leakage of the to-be-processed image caused by the target image.
  • the electronic device when a user shares an image, the electronic device automatically erases the location information of the image, shooting data information, and information such as the model of the electronic device, thereby protecting the network environment where the image is shared.
  • this technical solution mainly hides the content such as the network environment in which the image is shared, and does not protect the content of the image itself.
  • the criminals can easily match the relevant information of the characters and draw the relationship between the characters based on the selfie image shared by the user, combined with face recognition models such as FaceNet and massive information on the Internet. Network, so as to achieve the profit behavior of "human flesh” or "leaving information" to users. That is to say, the related art has the problem that the privacy of the selfie image shared by the user cannot be well protected.
  • the image processing method of the embodiment of the present application by performing noise processing on the image to be processed, the information of the processed target image is not easily recognized by human eyes or machines, so that when the target image is shared, the user's information can be better protected. privacy. It can be understood that the core idea of protecting image privacy in the embodiments of the present application is also applicable to language protection or other privacy protection scenarios.
  • the image to be processed may include user privacy such as human facial features and fingerprints.
  • the preset noise model can generate noise according to the pre-trained noise data, and add the noise to the position corresponding to the user's privacy (for example, facial features) in the image to be processed, thereby generating the target image.
  • the user when adding noise to the image to be processed, can adjust the degree of "invisibility" of the image to be processed or the degree of "distortion" of the image to be processed.
  • the similarity indicates the degree of similarity between the target image and the image to be processed. The greater the similarity, the more similar the target image and the image to be processed; the smaller the similarity, the less similar the target image and the image to be processed.
  • the first similarity between the target image and the image to be processed in the first space is greater than the first preset processing similarity. It can be understood that the first similarity between the target image and the image to be processed in the first space is larger, and the target image It is similar to the image to be processed in the first space.
  • the second similarity between the target image and the image to be processed in the second space is less than the second preset processing similarity. It can be understood that the second similarity between the target image and the image to be processed in the second space is smaller, and the target image It is quite different from the image to be processed in the second space.
  • the original space can be understood as the space viewed from the perspective of the human eye.
  • the feature space can be understood as the space “viewed” from the perspective of the machine.
  • the machine can obtain the image features of the image to be processed.
  • the image features generally include the high-dimensional abstract features of the image to be processed, which can better reflect the image to be processed. details, etc.
  • Electronic devices may include smart phones, digital cameras, tablet computers, or other terminal devices with photographing functions.
  • the first space is the original space
  • the second space is the feature space
  • the first similarity between the target image and the image to be processed in the original space is greater than the first preset processing similarity
  • the second similarity between the target image and the image to be processed in the feature space is smaller than the second preset processing similarity.
  • the first space is the feature space
  • the second space is the original space
  • the first similarity between the target image and the image to be processed in the feature space is greater than the first preset processing similarity
  • the second similarity between the target image and the image to be processed in the original space is smaller than the second preset processing similarity.
  • the image processing method further includes:
  • 050 Obtain the first training similarity between the training image and the noise-added image in the first space
  • 060 Obtain the second training similarity between the training image and the noise-added image in the second space
  • 070 Adjust the noise model to be trained according to the first training similarity and the second training similarity, so that the adjusted first training similarity is greater than the first preset training similarity, and the adjusted second training similarity is smaller than the second training similarity.
  • the training similarity is preset and the adjusted noise model to be trained is used as the preset noise model.
  • the image processing method of the above-mentioned embodiment can be implemented by the image processing apparatus 100 of the embodiment of the application itself.
  • the image processing apparatus 100 further includes a second acquisition module 30 , a second noise addition module 40 , a third acquisition module 50 , a fourth acquisition module 60 and an adjustment module 70 .
  • the second acquisition module 30 is used for acquiring training images.
  • the second noise adding module 40 is configured to perform noise processing on the training image by using the noise model to be trained to obtain a noise added image.
  • the third obtaining module 50 is configured to obtain the first training similarity between the training image and the noise-added image in the first space.
  • the fourth obtaining module 60 is configured to obtain the second training similarity between the training image and the noise-added image in the second space.
  • the adjustment module 70 is configured to adjust the noise model to be trained according to the first training similarity and the second training similarity, so that the adjusted first training similarity is greater than the first preset training similarity and the adjusted second training similarity. is less than the second preset training similarity, and the adjusted noise model to be trained is used as the preset noise model.
  • the image processing method of the above-mentioned embodiment can be implemented by the electronic device 200 of the embodiment of the application itself.
  • the processor 201 is used for acquiring a training image; and for performing noise processing on the training image by using the noise model to be trained to obtain a noise-added image; and for acquiring the first space of the training image and the noise-added image in the first space a training similarity; and a second training similarity for obtaining the training image and the noise-added image in the second space; and for adjusting the noise model to be trained according to the first training similarity and the second training similarity, so that The adjusted first training similarity is greater than the first preset training similarity, the adjusted second training similarity is less than the second preset training similarity, and the adjusted noise model to be trained is used as the preset noise model.
  • noise data that can effectively protect the privacy of the training image is obtained, and a preset noise model is generated according to the noise data.
  • the preset noise model can be used to process the image.
  • the image is subjected to noise processing to obtain a target image that hides the user's privacy in the image to be processed.
  • the training images may include user privacy such as human facial features and fingerprints.
  • the noise model to be trained can generate noise, and add the generated noise to the position corresponding to the user's privacy in the training image, thereby generating a noise-added image.
  • the adjusted first training similarity is greater than the first preset training similarity. It can be understood that the adjusted noise-added image and the training image have a larger first training similarity in the first space, and the adjusted noise-added image has a larger first training similarity. It is similar to the training image in the first space, and there will be no major difference between human eyes.
  • the adjusted second training similarity is less than the second preset training similarity. It can be understood that the adjusted noise-added image and the training image have a larger second training similarity in the second space, and the adjusted noise-added image has a larger second training similarity. and the training images are quite different in the second space.
  • step 050 includes:
  • 051 Obtain the first loss of the training image and the noise-added image in the first space, wherein the size of the first loss is negatively correlated with the size of the first training similarity;
  • the image processing method also includes:
  • Step 060 includes:
  • 061 Obtain a second loss of the noise-added image and the reference image in the second space, wherein the size of the second loss is negatively correlated with the size of the second training similarity;
  • Step 070 includes:
  • 072 Adjust the noise model to be trained according to the target loss so that the target loss is smaller than the preset loss, and use the adjusted noise model to be trained as the preset noise model.
  • the third obtaining module 50 includes a first obtaining unit 51 .
  • the first obtaining unit 51 is configured to obtain the first loss of the training image and the noise-added image in the first space, wherein the size of the first loss is negatively correlated with the size of the first training similarity.
  • the image processing apparatus 100 further includes a fifth acquisition module 80 .
  • the fifth acquisition module 80 is used to acquire a reference image, the similarity between the reference image and the training image in the first space is greater than the first preset reference similarity, and the similarity between the reference image and the training image in the second space is smaller than the second preset. Set reference similarity.
  • the fourth acquisition module 60 includes a second acquisition unit 61 .
  • the second obtaining unit 61 is configured to obtain the second loss of the noise-added image and the reference image in the second space, wherein the magnitude of the second loss is negatively correlated with the magnitude of the second training similarity.
  • the adjustment module 70 includes a determination unit 71 and an adjustment unit 72 .
  • the determining unit 71 is used to determine the target loss according to the first loss and the second loss; the adjusting unit 72 is used to adjust the noise model to be trained according to the target loss so that the target loss is less than the preset loss and the adjusted noise model to be trained is used as the preset noise model.
  • the image processing method of the above-mentioned embodiment can be implemented by the electronic device 200 of the embodiment of the application itself.
  • the processor 201 is used to obtain the first loss of the training image and the noise-added image in the first space, wherein the size of the first loss is negatively correlated with the size of the first training similarity; and used to obtain the reference image, the similarity between the reference image and the training image in the first space is greater than the first preset reference similarity, and the similarity between the reference image and the training image in the second space is less than the second preset reference similarity; and for obtaining a second loss of the noised image and the reference image in the second space, wherein the magnitude of the second loss is negatively correlated with the magnitude of the second training similarity; the target loss is determined according to the first loss and the second loss; and The noise model to be trained is adjusted according to the target loss so that the target loss is smaller than the preset loss, and the adjusted noise model to be trained is used as the preset noise model.
  • the noise model to be trained is adjusted so that the noise-added image and the reference image are similar in the second space, so as to better hide the user's privacy.
  • the size of the first loss is negatively correlated with the size of the first training similarity. It can be understood that the smaller the first loss is, the greater the first training similarity is, and the training image and the noised image are The more similar in the first space.
  • the reference image may include user privacy such as human facial features and fingerprints, and the user privacy in the reference image and the user privacy in the training image are from different users.
  • the users in the training images are Asian users
  • the users in the reference images are European and American users.
  • the similarity between the reference image and the training image in the first space is greater than the first preset reference similarity. similar in space.
  • the similarity between the reference image and the training image in the second space is less than the second preset reference similarity. It can be understood that the similarity between the reference image and the training image in the second space is smaller, and the reference image and the training image are in the second There is a big difference in space.
  • step 061 the size of the second loss is negatively correlated with the size of the second training similarity. It can be understood that the smaller the second loss is, the greater the second training similarity is, and the noised image and the reference image are in the second more similar in space.
  • the first loss and the second loss are weighted and summed to obtain the target loss.
  • the target loss will be reduced accordingly.
  • the target loss is reduced to less than the preset loss, it can be considered that the first loss and the second loss are both small, that is to say, the first training similarity is large, and the training image and the noise-added image are in the first space.
  • the second training similarity is also larger, at this time, the image features of the noise-added image in the second space deviate from the image features of the reference image, and the image features of the noise-added image in the second space deviate from the image features of the training image.
  • the noise model to be trained at this time is used.
  • the preset noise model when the user wants to share the to-be-processed image, the to-be-processed image can be directly subjected to noise processing according to the preset noise model, so as to generate a target image that hides the user's privacy.
  • step 072 preset a preset loss and preset adjustment times, adjust the noise model to be trained according to the target loss and the optimizer (Adam), when the target loss is less than the preset loss and/or the number of times to adjust the noise model to be trained reaches the preset
  • the adjustment times are set, the adjustment of the to-be-trained noise model is stopped, and the last-adjusted to-be-trained noise model is used as the preset noise model.
  • the preset loss is set to 0.8.
  • the adjustment of the noise model to be trained is stopped, and the last adjusted noise model to be trained is used as the preset noise model.
  • the first space is the original space
  • the first loss includes quality loss
  • step 051 includes:
  • 0511 Obtain the first loss using an algorithm among Mean Square Error (MSE), Peak Signal to Noise Ratio (PSNR), and Structural SIMilarity (SSIM).
  • MSE Mean Square Error
  • PSNR Peak Signal to Noise Ratio
  • SSIM Structural SIMilarity
  • the image processing method of the above-mentioned embodiment can be implemented by the image processing apparatus 100 of the embodiment of the application itself.
  • the first obtaining unit 51 includes a first obtaining sub-unit 511 .
  • the first obtaining subunit 511 is configured to obtain the first loss by using an algorithm among MSE, PSNR and SSIM.
  • the image processing method of the above-mentioned embodiment can be implemented by the electronic device 200 of the embodiment of the application itself.
  • the processor 201 is configured to obtain the first loss by using an algorithm among MSE, PSNR and SSIM.
  • the first loss of the training image and the noise-added image in the first space can be better calculated by an algorithm, so that the difference between the training image and the noise-added image in the first space can be measured.
  • the MSE and PSNR algorithms can better calculate the difference between the training image and the noise-added image in the first space.
  • the training image and the noise-added image are color images
  • SSIM can better calculate the difference between the training image and the noise-added image in the first space.
  • the first loss is obtained using SSIM.
  • the first loss 1-SSIM(x, y), where SSIM is the image quality loss function, x is the training image, and y is the noised image.
  • SSIM is also an indicator to measure the similarity between training images and noised images. This indicator defines structural information as independent of brightness and contrast from the perspective of image composition, reflects the properties of object structure in the scene, and models distortion as brightness A combination of three different factors: , contrast and structure.
  • the mean is used as an estimate of brightness
  • the standard deviation is used as an estimate of contrast
  • the covariance is used as a measure of structural similarity.
  • the value range of SSIM is (0, 1).
  • the second space is a feature space
  • the second loss includes a feature loss
  • step 061 includes:
  • 0612 Obtain a second loss based on image features.
  • the image processing method of the above-mentioned embodiment can be implemented by the image processing apparatus 100 of the embodiment of the application itself.
  • the second obtaining unit 61 includes a first determining subunit 611 and a second obtaining subunit 612 .
  • the first determination subunit 611 is used to determine the image features using various algorithms in FaceNet, InsightFace, SVM and DNN.
  • the second obtaining subunit 612 is configured to obtain the second loss according to the image feature.
  • the image processing method of the above-mentioned embodiment can be implemented by the electronic device 200 of the embodiment of the application itself.
  • the processor 201 is used for determining image features by using various algorithms in FaceNet, InsightFace, SVM and DNN; and for obtaining the second loss according to the image features.
  • a mainstream face recognition model is used to calculate the second loss, and the noise model to be trained is adjusted according to the second loss and the first loss to obtain a preset noise model, so that the target image obtained through the preset noise model is not easily used by mainstream people.
  • the face recognition model recognizes and protects the privacy of the user's face. It can be understood that FaceNet, InsightFace, SVM and DNN are all face recognition models.
  • the image feature that is, the Embedding feature
  • the image can be calculated by a neural network to generate a feature vector, which can generally include high-dimensional abstract features of the image, and can better reflect the detailed features of the image.
  • the intermediate image features corresponding to multiple different face recognition models can be obtained, and the multiple intermediate image features are spliced together to form the noise-added image or reference image.
  • image features include dimensions, and stitching together multiple intermediate image features includes: adding the dimensions of the multiple intermediate image features to obtain higher-dimensional image features.
  • one intermediate image feature is 4-dimensional
  • the other intermediate image feature is 2-dimensional
  • the dimension of the image features obtained by splicing is 6-dimensional.
  • the intermediate image features include dimensions, and the dimensions of multiple intermediate image features are the same, and splicing the multiple intermediate image features together includes: weighting the multiple intermediate image features to obtain image features of the same dimension.
  • one intermediate image feature is 2-dimensional
  • the other intermediate image feature is 2-dimensional. After weighting processing, the spliced image features have a 2-dimensional dimension.
  • the second loss is related to the similarity between the image features of the noised image and the image features of the reference image.
  • One of Manhattan distance, Euclidean distance, or cosine similarity may be used to characterize the similarity between the image feature of the noised image and the image feature of the reference image.
  • the cosine similarity method is used to calculate the similarity between the image feature of the noised image and the image feature of the reference image. In this way, the similarity can be calculated faster, so that the image features of the noise-added image deviate from the image features of the training image and the image features of the reference image more quickly.
  • the second loss 1-cos(y_embedding, z_embedding), where cos is the cosine similarity, y_embedding is the noised image, and z_embedding is the reference image.
  • Cosine similarity that is, by calculating the cosine of the angle between the image features of the noised image and the image features of the reference image to evaluate their similarity.
  • the range of the cosine value is between [-1, 1]. The closer the cosine value is to 1, the second loss is to 0, indicating that the image features of the noise-added image are closer to the image features of the reference image.
  • the second loss tends to 2, indicating that the image features of the noised image are more opposite to the image features of the reference image; the cosine value is close to 0, and the second loss tends to 1, indicating that the image of the noised image is The features are nearly orthogonal to the image features of the reference image.
  • step 051 when the first space is a feature space, the first loss includes feature loss, and step 051 includes: determining image features using various algorithms in FaceNet, InsightFace, SVM and DNN; and The second loss is obtained according to the image features; when the second space is the original space, the second loss includes quality loss, and step 061 includes: using an algorithm among MSE, PSNR and SSIM to obtain the first loss.
  • the noise-added image includes a first face region
  • the reference image includes a second face region. Before step 0611, it includes:
  • the image processing method of the above-mentioned embodiment can be implemented by the image processing apparatus 100 of the embodiment of the application itself.
  • the second acquiring unit 61 further includes a cropping subunit 613 and an alignment subunit 614 .
  • the cropping subunit 613 is used for cropping the noise-added image and the reference image respectively to obtain the first face region and the second face region.
  • the alignment subunit 614 is used for aligning the first face region and the second face region.
  • the image processing method of the above-mentioned embodiment can be implemented by the electronic device 200 of the embodiment of the application itself.
  • the processor 201 is used for cropping the noise-added image and the reference image to obtain the first face region and the second face region; and for aligning the first face region and the second face region.
  • the face area can be cropped from the noise-added image and the reference image, and aligned into images of the same size, that is, the face input data of a fixed size is formed, so that the face recognition model can obtain the same size after processing.
  • the image features of the noisy image and the image features of the reference image are formed from the noise-added image and the reference image, and aligned into images of the same size, that is, the face input data of a fixed size is formed, so that the face recognition model can obtain the same size after processing.
  • the version-RFB model is used to crop and align the first face region and the second face region.
  • version-RFB is a lightweight face detection model.
  • the version-RFB model can determine whether there is a face area in the image, and at the same time, it can identify, frame and even crop out the face area in the image, and the version-RFB model can maintain a cropping rate of 98%. Lightweight.
  • the training image and the noise-added image include a first text region
  • the reference image includes a second text region
  • a first loss of the training image and the noise-added image in the first space is obtained
  • the noise-added image and the reference image are cropped respectively image to obtain the first text area and the second text area; align the first text area and the second text area; use the correlation model to determine the text features corresponding to the first text area and the second text area
  • the second loss of the reference image in the second space the target loss is determined according to the first loss and the second loss
  • the noise model to be trained is adjusted according to the target loss so that the target loss is smaller than the preset loss and the adjusted noise model to be trained is used as Default noise model. In this way, the privacy of the text can be protected.
  • step 080 includes:
  • the fifth acquisition module 80 includes a first selection unit 81 and a second selection unit 82 .
  • the first selection unit 81 is used for selecting images to be selected with a preset number of frames from the preset image set according to the training images, and the similarity of each frame of the to-be-selected image and the training image in the first space is greater than the first preset reference similarity.
  • the second selection unit 82 is configured to select the image to be selected with the smallest similarity with the training image in the second space as the reference image, so that the similarity between the reference image and the training image in the second space is smaller than the second preset reference similarity .
  • the image processing method of the above-mentioned embodiment can be implemented by the electronic device 200 of the embodiment of the application itself.
  • the processor 201 is configured to select an image to be selected with a preset number of frames from a preset image set according to the training image, and the similarity of each frame of the to-be-selected image and the training image in the first space is greater than the similarity of the first preset reference and selecting the image to be selected with the smallest similarity with the training image in the second space as the reference image, so that the similarity between the reference image and the training image in the second space is smaller than the second preset reference similarity.
  • the reference image will be used as the basis for the offset of the image features of the noised image.
  • the image features of the noised image in the second space are biased towards the image features of the reference image, while deviating from the image features of the training image. , in order to achieve the effect of protecting user privacy in training images.
  • the preset image set includes multiple frames of images to be selected, and the images to be selected include a face region.
  • the similarity between the image to be selected and the training image in the first space is greater than the similarity of the first preset reference. It can be understood that the similarity between the image to be selected and the training image in the first space is greater, and the image to be selected and the training image have a greater similarity.
  • the facial features of the human face are relatively similar.
  • the similarity between the reference image and the training image in the second space is less than the second preset reference similarity. It can be understood that the similarity between the reference image and the training image in the second space is smaller, and the reference image and the training image are in the second
  • the image features in space are quite different.
  • the preset number of frames is 10 frames. In this way, it is beneficial to obtain more images to be selected that are similar to the training images in the first space but are quite different from the training images in the second space. It can be understood that, in other implementation manners, the preset number of frames may also be 5 frames, 15 frames or other values, which are not limited herein.
  • step 071 includes:
  • 0712 Determine the target loss according to the first loss, the first weight, the second loss and the second weight.
  • the determination unit 71 includes a second determination subunit 711 and a third determination subunit 712 .
  • the second determination subunit 711 is configured to determine the first weight of the first loss and the second weight of the second loss, respectively.
  • the third determination subunit 712 is configured to determine the target loss according to the first loss, the first weight, the second loss and the second weight.
  • the image processing method of the above-mentioned embodiment can be implemented by the electronic device 200 of the embodiment of the application itself.
  • the processor 201 is configured to respectively determine the first weight of the first loss and the second weight of the second loss; and to determine according to the first loss, the first weight, the second loss and the second weight target loss.
  • the target image and the image to be processed are more similar in the first space, and the difference between the target image and the image to be processed in the second space is relatively small;
  • the first weight is less than the second weight, for example, the first weight is 0.1 and the second weight is 0.9, the proportion of the first loss in the target loss is smaller than the proportion of the second loss in the target loss, this When focusing more on the "stealth" effect of the second space, that is, the target image and the image to be processed are similar in the first space, while the difference between the target image and the image to be processed in the second space is relatively large.
  • the computer-readable storage medium of the embodiment of the present application stores a computer program.
  • the program is executed by the processor, the steps of the image processing method of the above-described embodiment are realized.
  • 020 Perform noise processing on the image to be processed by using a preset noise model to obtain a target image, where the first similarity between the target image and the image to be processed in the first space is greater than the first preset similarity for processing, and the target image and the image to be processed have a first similarity greater than the first preset similarity.
  • the second similarity in the second space is smaller than the second preset processing similarity, where the first space is one of the original space and the feature space, and the second space is the other of the original space and the feature space.
  • the computer-readable storage medium can be set in the electronic device 200 or in a cloud server, and the electronic device 200 can communicate with the cloud server to obtain a corresponding program.
  • a computer program includes computer program code.
  • the computer program code may be in source code form, object code form, an executable file or some intermediate form, or the like.
  • Computer-readable storage media may include: any entity or device capable of carrying computer program codes, recording media, U disk, removable hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random storage Access memory (RAM, Random Access Memory), and software distribution media, etc.
  • the processor may refer to the processor 201 included in the electronic device 200 .
  • the processor 201 can be a central processing unit (Central Processing Unit, CPU), and can also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSPs), application specific integrated circuits (Application Specific Integrated Circuits, ASIC), ready-made processors.
  • first and second are only used for descriptive purposes, and should not be construed as indicating or implying relative importance or implying the number of indicated technical features. Thus, features delimited with “first”, “second” may expressly or implicitly include at least one of said features. In the description of the present application, “plurality” means at least two, such as two, three, unless expressly and specifically defined otherwise.
  • any description of a process or method in the flowcharts or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing a specified logical function or step of the process , and the scope of the preferred embodiments of the present application includes alternative implementations in which the functions may be performed out of the order shown or discussed, including performing the functions substantially concurrently or in the reverse order depending upon the functions involved, which should It is understood by those skilled in the art to which the embodiments of the present application belong.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Computer Security & Cryptography (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Bioethics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

La présente invention concerne un procédé de traitement d'image, un appareil de traitement d'image (100), un dispositif électronique (200) et un support d'enregistrement. Le procédé de traitement d'image comprend les étapes consistant à : acquérir une image à traiter ; et effectuer, à l'aide d'un modèle de bruit prédéfini, un traitement d'ajout de bruit sur l'image à traiter de façon à obtenir une image cible, une première similarité de l'image cible et de l'image à traiter dans un premier espace étant supérieure à une première similarité de traitement prédéfinie, une seconde similarité de l'image cible et de l'image à traiter dans un second espace étant inférieure à une seconde similarité de traitement prédéfinie, le premier espace étant l'un d'un espace d'origine et d'un espace de caractéristiques, et le second espace étant l'autre de l'espace d'origine et de l'espace de caractéristiques.
PCT/CN2021/079485 2021-03-08 2021-03-08 Procédé de traitement d'image, appareil de traitement d'image, dispositif électronique et support d'enregistrement WO2022187991A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202180088292.7A CN116964582A (zh) 2021-03-08 2021-03-08 图像处理方法、图像处理装置、电子设备及存储介质
PCT/CN2021/079485 WO2022187991A1 (fr) 2021-03-08 2021-03-08 Procédé de traitement d'image, appareil de traitement d'image, dispositif électronique et support d'enregistrement

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/079485 WO2022187991A1 (fr) 2021-03-08 2021-03-08 Procédé de traitement d'image, appareil de traitement d'image, dispositif électronique et support d'enregistrement

Publications (1)

Publication Number Publication Date
WO2022187991A1 true WO2022187991A1 (fr) 2022-09-15

Family

ID=83227274

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/079485 WO2022187991A1 (fr) 2021-03-08 2021-03-08 Procédé de traitement d'image, appareil de traitement d'image, dispositif électronique et support d'enregistrement

Country Status (2)

Country Link
CN (1) CN116964582A (fr)
WO (1) WO2022187991A1 (fr)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110163218A (zh) * 2019-04-10 2019-08-23 阿里巴巴集团控股有限公司 基于图像识别的脱敏处理方法以及装置
US20200082147A1 (en) * 2018-09-06 2020-03-12 Idemia Identity & Security France Method of extracting features from a fingerprint represented by an input image
CN111310734A (zh) * 2020-03-19 2020-06-19 支付宝(杭州)信息技术有限公司 保护用户隐私的人脸识别方法和装置
CN111783146A (zh) * 2020-09-04 2020-10-16 支付宝(杭州)信息技术有限公司 基于隐私保护的图像处理方法、装置和电子设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200082147A1 (en) * 2018-09-06 2020-03-12 Idemia Identity & Security France Method of extracting features from a fingerprint represented by an input image
CN110163218A (zh) * 2019-04-10 2019-08-23 阿里巴巴集团控股有限公司 基于图像识别的脱敏处理方法以及装置
CN111310734A (zh) * 2020-03-19 2020-06-19 支付宝(杭州)信息技术有限公司 保护用户隐私的人脸识别方法和装置
CN111783146A (zh) * 2020-09-04 2020-10-16 支付宝(杭州)信息技术有限公司 基于隐私保护的图像处理方法、装置和电子设备

Also Published As

Publication number Publication date
CN116964582A (zh) 2023-10-27

Similar Documents

Publication Publication Date Title
CN108764091B (zh) 活体检测方法及装置、电子设备和存储介质
US11551377B2 (en) Eye gaze tracking using neural networks
KR102324706B1 (ko) 얼굴인식 잠금해제 방법 및 장치, 기기, 매체
US11263441B1 (en) Systems and methods for passive-subject liveness verification in digital media
US9307221B1 (en) Settings of a digital camera for depth map refinement
JP6267224B2 (ja) 最良の写真を検出及び選択する方法及びシステム
US8184870B2 (en) Apparatus, method, and program for discriminating subjects
US7720302B2 (en) Method, apparatus and program for image processing
US7848545B2 (en) Method of and system for image processing and computer program
US20190325564A1 (en) Image blurring methods and apparatuses, storage media, and electronic devices
US20210279469A1 (en) Image signal provenance attestation
US20050249429A1 (en) Method, apparatus, and program for image processing
US11004179B2 (en) Image blurring methods and apparatuses, storage media, and electronic devices
KR101725884B1 (ko) 이미지들에 대한 자동 프로세싱
US11373449B1 (en) Systems and methods for passive-subject liveness verification in digital media
US20200218886A1 (en) Facial Anti-Spoofing Method Using Variances In Image Properties
WO2016197657A1 (fr) Procédé et appareil de traitement de photographie
WO2018133322A1 (fr) Procédé de mise au point automatique à détection de phase et dispositif pour cela
CN107368817B (zh) 人脸识别方法和装置
CN113822927A (zh) 一种适用弱质量图像的人脸检测方法、装置、介质及设备
WO2022187991A1 (fr) Procédé de traitement d'image, appareil de traitement d'image, dispositif électronique et support d'enregistrement
Chen et al. Hybrid saliency detection for images
US20230419737A1 (en) Methods and systems for detecting fraud during biometric identity verification
US11470388B2 (en) Method and apparatus for preventing forgery of data, method and apparatus for detecting forgery of data
Cao et al. Forensic detection of noise addition in digital images

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21929474

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 202180088292.7

Country of ref document: CN

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21929474

Country of ref document: EP

Kind code of ref document: A1