Video classification privacy protection method, system and device and storage medium

By encrypting video images and performing inverse transformation using a classification model trained with the same key information, the problem of insufficient privacy protection in video classification is solved, achieving a balance between information security and model performance during transmission.

CN119052389APending Publication Date: 2024-11-29CHINA MOBILE GRP BEIJING +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411190558.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Existing technologies have insufficient privacy protection in the video classification process, leading to the risk of data leakage during transmission. Furthermore, existing privacy protection measures may affect model performance or cause response delays.

Method used

Encrypted images are generated by encrypting video images, and then a classification model trained with the same key information is used for inverse transformation. Classification and recognition are then performed directly on the server side, avoiding decryption operations.

Benefits of technology

It achieves the protection of information security during transmission, shortens response time, maintains model classification accuracy and performance, and improves security.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119052389A_ABST
    Figure CN119052389A_ABST
Patent Text Reader

Abstract

The invention provides a video classification privacy protection method, system and device and a storage medium, and relates to the technical field of video classification, and the method comprises the steps: obtaining an encrypted image which is generated by carrying out the first conversion processing of a video image; inputting the encrypted image into a classification model to obtain a classification result; wherein the classification model is obtained by performing second conversion processing on the original model, the first conversion processing and the second conversion processing use the same key information, and the classification model is used for performing inverse conversion and classification identification processing on the encrypted image. The encrypted images are input into the classification model, the encrypted images do not need to be decrypted, the calculation amount is reduced, the response time is shortened, decryption is not needed, the images on the server side still have the privacy protection effect, safety is improved, the classification model classifies the encrypted images obtained by using the same secret key information, and the classification efficiency is improved. The classification accuracy of the model is maintained, and the performance of the model is maintained while privacy protection is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of video classification, and in particular to a video classification privacy protection method, system, device and storage medium. Background Art

[0002] As artificial intelligence is widely used in video classification, it can realize functions such as face recognition, action recognition, and behavior analysis for video images. However, the video classification process involves biometric features. For example, face recognition relies on facial features, and the function implementation relies on personalized data. In the process of uploading videos to the server for video classification, there are privacy protection issues, such as leakage during the transmission process, model leakage, or leakage to the server operator.

[0003] In the prior art, some methods degrade video images, such as reducing resolution, scrambling, adding noise, etc., to avoid leakage during transmission. However, the degradation of video images will affect the classification accuracy of the model and reduce model performance. Another method is to encrypt the video image before transmission, decrypt it after transmission to the server, and then input it into the model for processing. However, due to the large amount of calculation required for encryption and decryption, the processing takes a long time, which may cause response delays, and the video image decrypted on the server side still has the risk of leakage. Summary of the invention

[0004] The present invention provides a video classification privacy protection method, system, device and storage medium, which are used to solve the defects of implementing privacy protection in the video image classification process in the prior art, such as affecting model performance, response delay and server-side leakage.

[0005] The present invention provides a video classification privacy protection method, comprising: Acquire an encrypted image, where the encrypted image is generated by performing a first conversion process on the video image; Inputting the encrypted image into a classification model to obtain a classification result; The classification model is obtained by performing a second conversion process on the original model, the first conversion process and the second conversion process use the same key information, and the classification model is used to perform inverse conversion and classification recognition processing on the encrypted image.

[0006] According to a video classification privacy protection method provided by the present invention, the classification model is obtained by the following method, including: Acquire an original model and key information, wherein the original model includes an image segmentation layer, a first linear transformation coding layer, and a classification processing layer; The first linear conversion coding layer is adjusted according to the key information to form a second linear conversion coding layer, and the second linear conversion coding layer performs an inverse conversion process corresponding to the first conversion process on the input data; The classification model is generated according to the image segmentation layer, the second linear transformation coding layer and the classification processing layer.

[0007] Among them, the original model is obtained by model training, the image segmentation layer is used to segment the input image according to a preset unit image size to form image block information, the first linear transformation coding layer is used to linearly transform and encode the image block information to form encoding information, and the classification processing layer is used to generate a classification result based on the encoding information.

[0008] According to a video classification privacy protection method provided by the present invention, the key information includes a first key, a second key and a third key; the first linear transformation coding layer includes a first position embedding matrix and a first block coding matrix; the first linear transformation coding layer is adjusted according to the key information to form a second linear transformation coding layer, including: generating a first permutation matrix according to the first key, and generating a second permutation matrix according to the second key; Adjust the second permutation matrix according to the third key to generate a third permutation matrix; adjusting the first position embedding matrix according to the first permutation matrix, so as to adjust the position information corresponding to each image block in the process of embedding the image block information into the position information, and generate a second position embedding matrix; The first block encoding matrix is ​​adjusted according to the third permutation matrix to adjust the pixel order in each image block and flip the pixels during the encoding of the image block information, thereby generating a second block encoding matrix.

[0009] According to a video classification privacy protection method provided by the present invention, the encrypted image is generated by the following method, including: According to a preset unit image size, the video image is segmented to obtain a plurality of unit image blocks; Generate linear conversion information according to preset key information; The unit image block is processed according to the linear conversion information to obtain the encrypted image.

[0010] According to a video classification privacy protection method provided by the present invention, the key information includes a first key, a second key and a third key; the linear conversion information is generated according to the preset key information, including: Generate a corresponding first vector according to the first key, generate a corresponding second vector according to the second key, and generate a corresponding third vector according to the third key; The linear conversion information is formed according to the first vector, the second vector, and the third vector.

[0011] According to a video classification privacy protection method provided by the present invention, the unit image block is processed according to the linear conversion information to obtain the encrypted image, including: Converting the unit image block into a one-dimensional vector to form a corresponding unit image vector; Arranging the order of each of the unit image vectors according to the first vector to obtain a first image vector sequence; Arranging the pixel value order of each of the unit image vectors in the first image vector sequence according to the second vector to obtain a second image vector sequence; According to the third vector, pixel value flipping processing is performed on the second image vector sequence to obtain the encrypted image.

[0012] The present invention also provides a video classification privacy protection system, comprising: a classification model generation subsystem, an image conversion subsystem and a classification model subsystem; The classification model generation subsystem is used to perform model training using unencrypted images to obtain an original model and to perform a second conversion process on the original model according to key information to generate a classification model; The image conversion subsystem is used to perform a first conversion process on the video image according to the key information to generate an encrypted image; The classification model subsystem is used to obtain the classification model from the classification model generation subsystem and input the encrypted image into the classification model to obtain a classification result; Among them, the classification model generation subsystem, the image conversion subsystem and the classification model subsystem cooperate with each other to execute the above-mentioned video classification privacy protection method.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a video classification privacy protection method as described in any one of the above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a video classification privacy protection method as described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements any of the above-mentioned video classification privacy protection methods.

[0016] The video classification privacy protection method, system, device and storage medium provided by the present invention have at least the following beneficial effects: The video image is first converted according to the key information to generate an encrypted image. The encrypted image can ensure information security and avoid leakage during transmission. After obtaining the encrypted image, the server inputs the encrypted image into the classification model. The classification model is obtained by performing a second conversion on the original model after training with the same key information, so that the classification model will perform a reverse conversion on the encrypted image during the classification and recognition process of the input encrypted image, thereby obtaining an accurate classification result. In this way, the encrypted image is directly input into the classification model, and there is no need to decrypt the encrypted image to reduce the amount of calculation, which is conducive to shortening the response time. Since there is no need to decrypt, the image on the server side still has a privacy protection effect, which is conducive to improving security. At the same time, the classification model classifies the encrypted image obtained using the same key information, has a normal classification result, maintains the classification accuracy of the model, and maintains the performance of the model while achieving privacy protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is one of the flow charts of a video classification privacy protection method provided by the present invention.

[0019] Figure 2 This is the second flow chart of a video classification privacy protection method provided by the present invention.

[0020] Figure 3 This is the third flow chart of a video classification privacy protection method provided by the present invention.

[0021] Figure 4 This is the fourth flow chart of a video classification privacy protection method provided by the present invention.

[0022] Figure 5 It is a schematic diagram of the processing process of a classification model in one embodiment of a video classification privacy protection method provided by the present invention.

[0023] Figure 6 It is a schematic diagram of a process of performing a first conversion process on a video image to generate an encrypted image in one embodiment of a video classification privacy protection method provided by the present invention.

[0024] Figure 7 It is a structural schematic diagram of a video classification privacy protection system provided by the present invention.

[0025] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0027] Combine the following Figure 1-Figure 4 A video classification privacy protection method of the present invention is described, comprising: S100: Acquire an encrypted image, where the encrypted image is generated by performing a first conversion process on a video image; S200: inputting the encrypted image into a classification model to obtain a classification result; The classification model is obtained by performing a second conversion process on the original model, the first conversion process and the second conversion process use the same key information, and the classification model is used to perform inverse conversion and classification recognition processing on the encrypted image.

[0028] The video image is first converted according to the key information to generate an encrypted image, which can ensure information security and avoid leakage during transmission. After obtaining the encrypted image, the server inputs the encrypted image into the classification model, which is obtained by performing a second conversion on the original model trained with the same key information, so that the classification model will perform a reverse conversion on the encrypted image during the classification and recognition process of the input encrypted image, thereby obtaining an accurate classification result.

[0029] In this way, the encrypted image is directly input into the classification model, and there is no need to decrypt the encrypted image to reduce the amount of calculation, which is beneficial to shorten the response time. Since there is no need for decryption, the image on the server side still has a privacy protection effect, which is beneficial to improve security. At the same time, the classification model classifies the encrypted image obtained using the same key information, has a normal classification result, maintains the classification accuracy of the model, and maintains the performance of the model while achieving privacy protection.

[0030] It should be noted that, in some embodiments of the present invention, the original model may be a VIT (Vision Transformer) model, which is obtained after model training using unencrypted video images. Figure 5 In the present invention, the VIT model realizes image classification and recognition. First, the input image is input to the image segmentation layer for segmentation to form multiple image blocks, and then the multiple image blocks are input to the linear transformation coding layer (corresponding to Figure 5 The image is linearly projected and Transformer Encoder is used to perform linear transformation and encoding, and the encoding result is input into the classification processing layer, that is, the fully connected multilayer perceptron (MLP) layer, to obtain the classification result of image recognition.

[0031] It should be emphasized that the present invention does not simply merge the decryption process into the model, but adjusts its processing parameters according to the processing flow of the original model to perform a second conversion process to form a classification model, and then the classification model realizes the inverse conversion of the encrypted image. In other words, the classification model does not add additional processing flows, but only changes the parameters during the processing process, which will not cause the classification model to perform additional calculation steps and will not cause a significant increase in the processing time of the classification model. For example, after the VIT model performs the second conversion process, it still executes the process of segmenting, linear conversion and encoding the input image, and MLP layer processing. The second conversion process can be to adjust the parameters in the linear conversion and encoding to achieve the inverse conversion.

[0032] It can be understood that the process of performing the first conversion processing on the video image according to the key information to obtain the encrypted image is related to the processing process of the classification model. For example, when the classification model is a VIT model, the VIT model includes linear conversion and encoding processing. The first conversion processing can be a linear encryption method so that the parameters of the original VIT model can be adjusted according to the key information, that is, the second conversion processing, so that the classification VIT model can achieve inverse conversion during linear conversion and encoding processing.

[0033] The first conversion process and the second conversion process are both performed according to the same key information, so that the classification model has a corresponding relationship with the encrypted image. The classification model only has high-performance and high-accuracy classification performance for the encrypted image formed using the same key information, and the classification model will not be leaked.

[0034] refer to Figure 2 and Figure 6 In some embodiments of a video classification privacy protection method of the present invention, the encrypted image is generated by the following method, including: S010: Segmenting the video image according to a preset unit image size to obtain a plurality of unit image blocks; S020: Generate linear conversion information according to preset key information; S030: Processing the unit image block according to the linear conversion information to obtain the encrypted image.

[0035] The video image is segmented based on the preset unit image size to obtain multiple unit image blocks, and the size of each unit image block is the same as the preset unit image size. For example, a 224×224×3 video image is segmented based on the preset unit image size of 16×16×3 to obtain 196 unit image blocks.

[0036] Linear transformation information is generated according to the key information, and the unit image block is encrypted and transformed using the linear transformation information to obtain an encrypted image. The encrypted image requires the key information to be restored. Even if the packet is captured or intercepted during the transmission process, the video image cannot be restored due to the lack of key information. Therefore, the privacy protection function of the transmission process can be realized.

[0037] Since the model's processing of input information usually involves multiple steps, for example, after the VIT model segments the input image, it will also perform linear mapping and position encoding on the segmented image, that is, linear transformation coding processing. In order to improve the reliability of the encrypted image, multiple encryption transformation steps can be performed on the video image accordingly.

[0038] For the above reasons, reference Figure 3 In some embodiments of a video classification privacy protection method of the present invention, the key information includes a first key, a second key and a third key; S020 includes: S021: Generate a corresponding first vector according to the first key, generate a corresponding second vector according to the second key, and generate a corresponding third vector according to the third key; S022: Form the linear conversion information according to the first vector, the second vector and the third vector.

[0039] By using the first key, the second key and the third key to generate the first vector, the second vector and the third vector respectively, it is convenient to perform multiple encryption steps on the video image subsequently, thereby increasing the difficulty of cracking the encrypted image and enhancing the reliability of the encrypted image.

[0040] It should be noted that the key and the generated vector are corresponding, that is, different keys will generate different vectors, and different keys will not generate the same vector.

[0041] refer to Figure 4 and Figure 6In some embodiments of a video classification privacy protection method of the present invention, the S030 includes: S031: converting the unit image block into a one-dimensional vector to form a corresponding unit image vector; S032: Arrange the order of each of the unit image vectors according to the first vector to obtain a first image vector sequence; S033: Arranging the pixel value order of each of the unit image vectors in the first image vector sequence according to the second vector to obtain a second image vector sequence; S034: Perform pixel value flipping processing on the second image vector sequence according to the third vector to obtain the encrypted image.

[0042] By converting the unit image block into a one-dimensional vector, it is convenient to simplify the calculation and improve the calculation processing efficiency, and each unit image block is converted into a corresponding unit image vector. The first vector generated by the first key is used to sort the unit image vectors, that is, to arrange them in units of unit image vectors to obtain a first image vector sequence. The second vector generated by the second key is used to further arrange the pixel values ​​in the unit image vector, that is, for each unit image vector in the first image vector sequence, its own pixel value is arranged based on the second vector, and the order of the unit image vector remains unchanged to obtain a second image vector sequence. The third vector generated by the third key is used to perform pixel flipping processing on the pixel values ​​in the second image vector sequence, that is, to determine the target pixel value based on the third vector and change the positive and negative of the target pixel value. In this way, triple encryption is achieved using the first vector, the second vector, and the third vector, which is conducive to improving the reliability of the encrypted image.

[0043] It is understandable that the second image vector sequence is formed into the third image vector sequence after the pixel flipping process. The image vector sequences can be arranged into the same size according to the size of the original video image to form an encrypted image for transmission.

[0044] In some embodiments of the present invention, the first, second, and third vectors are generated according to the first, second, and third keys. The HMAC algorithm may be used to perform a hash operation on the keys to obtain corresponding hash values, and the corresponding vectors are generated according to the hash values ​​combined with the length of the unit image vectors, so that the unit image vectors can be arranged and the pixels can be flipped accordingly according to the vectors.

[0045] In some embodiments of the present invention, the third vector may be a binary vector, that is, the third vector is composed of 0 and 1, and the length of the third vector is the same as the total length of the second image vector sequence, that is, each pixel value in the second image vector sequence will correspond to a 0 or 1 in the third vector. Accordingly, according to the corresponding 0 or 1, the following transformation is performed:

[0046] in, is the third image vector sequence, which is an intermediate state sequence; w is the width sequence number; h is the height sequence number; k is the pixel sequence number; is the second image vector sequence; is the kth value in the third vector.

[0047] After the above transformation, normalization conversion is performed:

[0048] in, is the sequence corresponding to the encrypted image; is the third image vector sequence, which is an intermediate state sequence; w is the width sequence number; h is the height sequence number; k is the pixel sequence number; is the second image vector sequence; is the kth value in the third vector.

[0049] Through the above process, pixel flipping processing is implemented based on the third vector.

[0050] The above is the process of video image encryption, which is usually performed on the sending device. The classification model side will be described later, which is usually performed on the receiving device, which can be a server, cloud device, etc.

[0051] In some embodiments of a video classification privacy protection method of the present invention, the classification model is obtained by the following method, including: Acquire an original model and key information, wherein the original model includes an image segmentation layer, a first linear transformation coding layer, and a classification processing layer; The first linear conversion coding layer is adjusted according to the key information to form a second linear conversion coding layer, and the second linear conversion coding layer performs an inverse conversion process corresponding to the first conversion process on the input data; The classification model is generated according to the image segmentation layer, the second linear transformation coding layer and the classification processing layer.

[0052] Among them, the original model is obtained by model training, the image segmentation layer is used to segment the input image according to a preset unit image size to form image block information, the first linear transformation coding layer is used to linearly transform and encode the image block information to form encoding information, and the classification processing layer is used to generate a classification result based on the encoding information.

[0053] The original model obtained through the pre-model training includes an image segmentation layer, a first linear conversion coding layer, and a classification processing layer. The parameters of the first linear conversion coding layer are adjusted by using the key information to form a second linear conversion coding layer, so that the second linear conversion coding layer performs linear mapping, encoding, and other processing on the input image information, and simultaneously implements the inverse processing corresponding to the first conversion processing. In this way, the classification model formed by the image segmentation layer, the second linear conversion coding layer, and the classification processing layer can accurately classify and identify the encrypted image generated using the same key information, and obtain accurate classification results.

[0054] There is no need to decrypt the encrypted image, and the encrypted image can be directly input into the classification model, which is beneficial to reduce the amount of calculation and shorten the response time. Since there is no need for decryption, the image on the server side still has a privacy protection effect, which is beneficial to improve security. At the same time, the classification model classifies the encrypted image obtained using the same key information, has a normal classification result, maintains the classification accuracy of the model, and maintains the performance of the model while achieving privacy protection.

[0055] refer to Figure 5 , in the VIT model, the image segmentation layer corresponds to Figure 5 The image is segmented in the linear transformation coding layer. Figure 5 In the figure, there are linear projections of the image and Transformer Encoder, and the classification processing layer corresponds to the MLP layer.

[0056] It should be emphasized that the original model itself includes a first linear transformation coding layer. Adjusting the parameters of the first linear transformation coding layer based on the key information will not add additional processing steps to the formed classification model, that is, it will not significantly increase the processing time of the model. It will only produce normal processing efficiency fluctuations based on the initial model and will not increase the data processing volume of the classification model.

[0057] In some embodiments of the present invention, the original model may be a VIT model, and the adjusted classification model also belongs to the VIT model.

[0058] In some embodiments of a video classification privacy protection method of the present invention, the key information includes a first key, a second key, and a third key; the first linear transformation coding layer includes a first position embedding matrix and a first block coding matrix; and the first linear transformation coding layer is adjusted according to the key information to form a second linear transformation coding layer, including: generating a first permutation matrix according to the first key, and generating a second permutation matrix according to the second key; Adjust the second permutation matrix according to the third key to generate a third permutation matrix; adjusting the first position embedding matrix according to the first permutation matrix, so as to adjust the position information corresponding to each image block in the process of embedding the image block information into the position information, and generate a second position embedding matrix; The first block encoding matrix is ​​adjusted according to the third permutation matrix to adjust the pixel order in each image block and flip the pixels during the encoding of the image block information, thereby generating a second block encoding matrix.

[0059] In the original model, the image segmentation layer segments the input image, which is the same as the preset unit image size in the process of obtaining the encrypted image, and the image segmentation layer also divides the image into the same image block information; the first linear transformation coding layer uses the first position embedding matrix and the first block coding matrix to process the input information to realize position information embedding, linear transformation coding, etc.; the classification processing layer classifies and identifies the encoded information to obtain the classification result.

[0060] The first key, the second key and the third key are used in the process of obtaining the encrypted image, and the first permutation matrix generated according to the first key is used to adjust the first position embedding matrix to form a second position embedding matrix, so that when the segmented image block information is processed by the second position embedding matrix, the order of the image block information is arranged, corresponding to the first vector sorting each unit image vector; the second permutation matrix is ​​generated according to the second key, and the second permutation matrix is ​​adjusted by the third key to form a third permutation matrix, and the first block coding matrix is ​​adjusted by the third permutation matrix to form a second block coding matrix, so that when the image block information is processed by the second coding matrix, the pixel values ​​in each image block are sorted and the pixels are flipped, corresponding to the second vector arranging each pixel value in the unit image vector and the third vector changing the positive and negative of the target pixel value.

[0061] In this way, the inverse processing of the encrypted image is achieved, so that the encoded information finally input to the classification processing layer is the same as the encoded information obtained by the classification processing layer when the unencrypted image is input to the original model. That is, for the classification processing layer, the unencrypted image is input to the original model and the encrypted image is input to the classification model after the second conversion processing. The two situations are the same, so the classification result output by the classification processing layer will not change. The classification model processes the encrypted image formed using the same key information, and the classification performance does not change, thereby maintaining the performance of the model while achieving privacy protection.

[0062] refer to Figure 7 ,The present invention also provides a video classification privacy protection system, including: a classification model generation subsystem, an image conversion subsystem and a classification model subsystem; The classification model generation subsystem is used to perform model training using unencrypted images to obtain an original model and to perform a second conversion process on the original model according to key information to generate a classification model; The image conversion subsystem is used to perform a first conversion process on the video image according to the key information to generate an encrypted image; The classification model subsystem is used to obtain the classification model from the classification model generation subsystem and input the encrypted image into the classification model to obtain a classification result; Among them, the classification model generation subsystem, the image conversion subsystem and the classification model subsystem cooperate with each other to execute the above-mentioned video classification privacy protection method.

[0063] The classification model generation subsystem can be a device that performs model training processing in advance, uses unencrypted images to train the initial model, obtains the original model, and then uses the key information to perform a second conversion process on the original model to form a classification model. The classification model is deployed to the classification model subsystem, which can be specifically in a server or cloud device. The image conversion subsystem can be a user device, a monitoring terminal device, etc., which obtains the same key information from the classification model generation subsystem, performs a first conversion process on the video image to form an encrypted image, and then uploads it to the classification model subsystem. The classification model subsystem uses the classification model to process the encrypted image to obtain a classification result. In this way, the classification model subsystem directly inputs the encrypted image into the classification model, and there is no need to decrypt the encrypted image to reduce the amount of calculation, which is conducive to shortening the response time. Since there is no need to decrypt, the encrypted image obtained still has a privacy protection effect, which is conducive to improving security. At the same time, the classification model classifies the encrypted image obtained using the same key information, has a normal classification result, maintains the classification accuracy of the model, and maintains the performance of the model while achieving privacy protection.

[0064] Figure 8An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the above-mentioned video classification privacy protection method.

[0065] An electronic device provided in the present invention may be a server, a cloud device, etc.

[0066] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0067] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a video classification privacy protection method provided by the above methods.

[0068] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute a video classification privacy protection method provided by the above methods.

[0069] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0070] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0071] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0072] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0073] All actions to obtain signals, information or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where they are located, and with the authorization given by the owner of the corresponding device.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A video classification privacy protection method, It is characterized in that include: Acquire an encrypted image, where the encrypted image is generated by performing a first conversion process on the video image; Inputting the encrypted image into a classification model to obtain a classification result; The classification model is obtained by performing a second conversion process on the original model, the first conversion process and the second conversion process use the same key information, and the classification model is used to perform inverse conversion and classification recognition processing on the encrypted image.

2. A video classification privacy protection method according to claim 1, It is characterized in that The classification model is obtained by the following methods, including: Acquire an original model and key information, wherein the original model includes an image segmentation layer, a first linear transformation coding layer, and a classification processing layer; The first linear conversion coding layer is adjusted according to the key information to form a second linear conversion coding layer, and the second linear conversion coding layer performs an inverse conversion process corresponding to the first conversion process on the input data; The classification model is generated according to the image segmentation layer, the second linear transformation coding layer and the classification processing layer. Among them, the original model is obtained by model training, the image segmentation layer is used to segment the input image according to a preset unit image size to form image block information, the first linear transformation coding layer is used to linearly transform and encode the image block information to form encoding information, and the classification processing layer is used to generate a classification result based on the encoding information.

3. A video classification privacy protection method according to claim 2, It is characterized in that The key information includes a first key, a second key, and a third key; the first linear transformation coding layer includes a first position embedding matrix and a first block coding matrix; and the step of adjusting the first linear transformation coding layer according to the key information to form a second linear transformation coding layer includes: generating a first permutation matrix according to the first key, and generating a second permutation matrix according to the second key; Adjust the second permutation matrix according to the third key to generate a third permutation matrix; adjusting the first position embedding matrix according to the first permutation matrix, so as to adjust the position information corresponding to each image block in the process of embedding the image block information into the position information, and generate a second position embedding matrix; The first block encoding matrix is ​​adjusted according to the third permutation matrix to adjust the pixel order in each image block and flip the pixels during the encoding of the image block information, thereby generating a second block encoding matrix.

4. A video classification privacy protection method according to claim 1, It is characterized in that The encrypted image is generated by the following method, including: According to a preset unit image size, the video image is segmented to obtain a plurality of unit image blocks; Generate linear conversion information according to preset key information; The unit image block is processed according to the linear conversion information to obtain the encrypted image.

5. A video classification privacy protection method according to claim 4, It is characterized in that The key information includes a first key, a second key and a third key; the generating linear conversion information according to the preset key information includes: Generate a corresponding first vector according to the first key, generate a corresponding second vector according to the second key, and generate a corresponding third vector according to the third key; The linear conversion information is formed according to the first vector, the second vector, and the third vector.

6. A video classification privacy protection method according to claim 5, It is characterized in that The step of processing the unit image block according to the linear conversion information to obtain the encrypted image includes: Converting the unit image block into a one-dimensional vector to form a corresponding unit image vector; Arranging the order of each of the unit image vectors according to the first vector to obtain a first image vector sequence; Arranging the pixel value order of each of the unit image vectors in the first image vector sequence according to the second vector to obtain a second image vector sequence; According to the third vector, pixel value flipping processing is performed on the second image vector sequence to obtain the encrypted image.

7. A video classification privacy protection system, It is characterized in that include: Classification model generation subsystem, image conversion subsystem and classification model subsystem; The classification model generation subsystem is used to perform model training using unencrypted images to obtain an original model and to perform a second conversion process on the original model according to key information to generate a classification model; The image conversion subsystem is used to perform a first conversion process on the video image according to the key information to generate an encrypted image; The classification model subsystem is used to obtain the classification model from the classification model generation subsystem and input the encrypted image into the classification model to obtain a classification result; The classification model generation subsystem, the image conversion subsystem and the classification model subsystem cooperate with each other to perform a video classification privacy protection method as described in any one of claims 1 to 6.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the video classification privacy protection method as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, a video classification privacy protection method as described in any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, a video classification privacy protection method as described in any one of claims 1 to 6 is implemented.