WO2024060666A1

WO2024060666A1 - Face image encryption/decryption method and apparatus, electronic device, and storage medium

Info

Publication number: WO2024060666A1
Application number: PCT/CN2023/097284
Authority: WO
Inventors: 石海林; 梅涛; 胡一博
Original assignee: 北京京东尚科信息技术有限公司
Priority date: 2022-09-22
Filing date: 2023-05-30
Publication date: 2024-03-28
Also published as: CN115601214A

Abstract

Provided are a face image encryption/decryption method and apparatus, an electronic device, a storage medium, a computer program product, and a computer program. The face image encryption method comprises: performing frequency domain processing on an original face image to obtain N frequency domain face images at different frequencies, wherein N is a positive integer; obtaining encryption seeds; performing convolutional encryption on the encryption seeds and the N frequency domain face images to obtain respective encrypted frequency domain face images of the N frequency domain face images; and performing frequency domain inverse processing on the N encrypted frequency domain face images to obtain an encrypted face image.

Description

Face image encryption/decryption method, device, electronic equipment and storage medium

Cross-references to related applications

This application claims priority from Chinese Patent Application No. 2022111562577 filed in China on September 22, 2022, the entire content of which is incorporated herein by reference.

Technical field

The present disclosure relates to the field of image processing technology, and specifically to a face image encryption/decryption method and its device, electronic equipment, storage media, computer program products and computer programs.

Background technique

In related technologies, encrypted face images can render most deep recognition models ineffective, achieve identity encryption, and prevent the encrypted images from undergoing changes that are perceptible to the human eye, ensuring normal sharing on social media. However, for common operations of uploading photos on social media, such as scaling and compressing, the encryption effect of the image will be affected, and the encrypted image cannot be restored to the original image, which has great limitations in practical applications.

Contents of the invention

The embodiments of the present disclosure provide a facial image encryption/decryption method and apparatus, electronic device, storage medium, computer program product, and computer program.

According to one aspect of an embodiment of the present disclosure, a method for encrypting a face image is provided, comprising:

Perform frequency domain processing on the original face image to obtain N frequency domain face images at different frequencies, where N is a positive integer;

Get encrypted seeds;

Perform convolution encryption processing on the encryption seed and N frequency domain face images to obtain the encrypted frequency domain face images of each of the N frequency domain face images;

Perform frequency domain inverse processing on N encrypted frequency domain face images to obtain encrypted face images.

In the embodiment of the present disclosure, after the original face image is converted into a frequency domain face image, encryption processing is performed based on the frequency domain face image. In this way, the transformation during the encryption process can be constrained to occur at edges that are difficult for the human eye to perceive, etc. high-frequency region, thus ensuring that the encryption is imperceptible and robust.

According to another aspect of the embodiment of the present disclosure, a method for decrypting a face image is provided, including:

Perform frequency domain processing on the encrypted face image to obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer;

Get the encryption key;

Perform convolution and decryption processing on the encryption key and N encrypted frequency domain face images to obtain frequency domain face images of each of the N encrypted frequency domain face images;

Perform frequency domain inverse processing on N frequency domain face images to obtain the original face image.

In the embodiment of the present disclosure, inverse processing is performed on the basis of face image encryption, which can decrypt the encrypted face image and restore the original face image. For the first time, the reversible model is applied to the face identity steganography task to achieve utilization A model efficiently implements encryption and decryption of face images.

According to one aspect of an embodiment of the present disclosure, there is provided a facial image encryption device, comprising:

The frequency domain processing module is used to perform frequency domain processing on the original face image and obtain N frequency domain face images at different frequencies, where N is a positive integer;

Obtain module, used to obtain encrypted seeds;

The encryption module is used to perform convolution encryption processing on the encryption seed and N frequency domain face images to obtain the encrypted frequency domain face images of each of the N frequency domain face images;

The frequency domain inverse processing module is used to perform frequency domain inverse processing on N encrypted frequency domain face images to obtain encrypted face images.

According to another aspect of the embodiment of the present disclosure, a device for decrypting facial images is provided, including:

The frequency domain processing module is used to perform frequency domain processing on encrypted face images and obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer;

Obtain module, used to obtain encryption keys;

The decryption module is used to perform convolution and decryption processing on the encryption key and N encrypted frequency domain face images to obtain frequency domain face images of each of the N encrypted frequency domain face images;

The frequency domain inverse processing module is used to perform frequency domain inverse processing on N frequency domain face images to obtain the original face image.

According to another aspect of the embodiment of the present disclosure, an electronic device is provided, including a memory and a processor;

Wherein, the processor reads the executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the encryption/decryption method of the face image according to any embodiment of the present disclosure.

According to another aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the face image encryption/decryption method of any embodiment of the present disclosure is implemented. .

According to another aspect of the embodiments of the present disclosure, a computer program product is provided, including a computer program. When executed by a processor, the computer program implements the encryption/decryption method of a facial image according to any embodiment of the present disclosure.

According to another aspect of an embodiment of the present disclosure, a computer program is provided, including a computer program code. When the computer program code is run on a computer, the computer performs the encryption of a face image according to any embodiment of the present disclosure. /decryption method.

It should be understood that what is described in this section is not intended to identify key or important features of the embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the present disclosure will become readily understood from the following description.

Description of drawings

Figure 1 is a flow chart of a facial image encryption method according to an embodiment of the present disclosure;

Figure 2 is a flow chart of a facial image encryption method according to an embodiment of the present disclosure;

Figure 3 is a flow chart of a facial image encryption method according to an embodiment of the present disclosure;

Figure 4 is a schematic diagram of a target face encryption network;

Figure 5 is a flow chart of a facial image encryption method according to an embodiment of the present disclosure;

Figure 6 is a flow chart of a facial image encryption method according to an embodiment of the present disclosure;

Figure 7 is a flow chart of a method for decrypting a face image according to an embodiment of the present disclosure;

Figure 8 is a flow chart of a method for decrypting a face image according to an embodiment of the present disclosure;

Figure 9 is a flow chart of a method for decrypting a face image according to an embodiment of the present disclosure;

Figure 10 is a schematic diagram of a target face decryption network;

Figure 11 is a flow chart of a facial image encryption method according to an embodiment of the present disclosure;

Figure 12 is a structural diagram of a face image encryption device according to an embodiment of the present disclosure;

Figure 13 is a structural diagram of a face image encryption device according to an embodiment of the present disclosure;

FIG14 is a structural diagram of a facial image decryption device according to an embodiment of the present disclosure;

Figure 15 is a structural diagram of a facial image decryption device according to an embodiment of the present disclosure;

Figure 16 is a block diagram of an electronic device used to implement the encryption/decryption method of a face image according to an embodiment of the present disclosure.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention and are not to be construed as limiting the present invention.

The following describes the facial image encryption/decryption method and its device, electronic device, storage medium, computer program product and computer program disclosed in the present invention in conjunction with the reference drawings.

Figure 1 is a flow chart of a face image encryption method according to an embodiment of the present disclosure. As shown in Figure 1, the method includes steps S11 to S14.

S11, perform frequency domain processing on the original face image, and obtain N frequency domain face images at different frequencies, where N is a positive integer.

The frequency of an image is an indicator of the intensity of grayscale changes in an image, so it can be considered that the edges of the image occupy the high-frequency band, while the main body of the image or the slowly changing grayscale area occupies the low-frequency band. Frequency domain processing of the original face image can transform the image in high-frequency areas such as edges that are difficult for the human eye to detect, thereby ensuring that the encryption is imperceptible and robust.

The frequency domain processing method of the image includes Fourier transform and wavelet transform. In the embodiment of the present disclosure, wavelet transform is used to perform frequency domain processing on the original face image to obtain N frequency domain face images at different frequencies, where N is positive integer. As a possible implementation method, wavelet transform is performed on the face image to obtain four branches, including one low-frequency face image and three high-frequency face images.

S12, obtain the encrypted seed.

The encryption seed is used to encrypt the frequency domain face image. At the same time, the encryption key of the original face image can also be obtained based on the encryption seed. The encryption key can be used to restore the encrypted face image to the original face image, realizing the human face image. Reversibility of face image encryption.

In some embodiments, the encryption seed is obtained by randomly sampling from a Gaussian normal distribution of the original face image.

S13: Perform convolution encryption processing on the encryption seed and N frequency domain face images to obtain encrypted frequency domain face images of each of the N frequency domain face images.

As a possible implementation method, construct an encryption model for face images, use the obtained encryption seeds and N frequency domain face images as network inputs, and perform convolution encryption processing through the model to obtain each of the N frequency domain face images. Encrypted frequency domain face images.

In some embodiments, an encryption model for face images can also be designed based on the traditional reversible model, and one model can be used to efficiently implement encryption and decryption of face images.

S14: Perform frequency domain inverse processing on the N encrypted frequency domain face images to obtain the encrypted face image.

In the aforementioned steps, the original face image was processed in the frequency domain to perform image transformation on high-frequency areas such as edges that are not easily perceived by the human eye. Correspondingly, after the encryption is completed, it is necessary to perform frequency domain inversion of the N encrypted frequency domain face images. Process to obtain the encrypted face image. The cosine similarity between the features of the encrypted face image and the original face image is low, but still maintains high visual quality, achieving traceless encryption of face identity.

In the embodiment of the present disclosure, frequency domain processing is performed on the original face image, N frequency domain face images at different frequencies are obtained, N is a positive integer, the encryption seed is obtained, and the encryption seed and N frequency domain face images are processed. Through convolution encryption processing, an encrypted frequency domain face image of each of the N frequency domain face images is obtained. The N encrypted frequency domain face images are subjected to frequency domain inverse processing to obtain an encrypted face image. In the embodiment of the present disclosure, after the original face image is converted into a frequency domain face image, encryption processing is performed based on the frequency domain face image. In this way, the transformation during the encryption process can be constrained to occur at edges that are difficult for the human eye to perceive, etc. high-frequency region, thus ensuring that the encryption is imperceptible and robust.

Figure 2 is a flow chart of a face image encryption method according to an embodiment of the present disclosure. As shown in Figure 2, the method includes steps S21 to S24.

S21, perform frequency domain processing on the original face image, and obtain N frequency domain face images at different frequencies, where N is a positive integer.

S22, obtain the encrypted seed.

For the introduction of steps S21 to S22, please refer to the relevant content records in the above embodiments, and will not be described again here.

S23. Perform convolution encryption processing on the encryption seed and N frequency domain face images to obtain the encryption key of the original face image and the encrypted frequency domain face images of each of the N frequency domain face images.

By performing convolution encryption processing on the encryption seed and N frequency domain face images, not only the encrypted frequency domain face images of each of the N frequency domain face images can be obtained, but also the encryption key of the original face image can be obtained. The encryption key can restore the encrypted face image to the original face image, achieving reversibility of face image encryption.

As a possible implementation method, combine the encryption seeds and N frequency domain face images in a set order to obtain a column vector to be encrypted, and treat the vector elements in the encrypted column vector row by row in positive order starting from the first row. , respectively perform convolution encryption processing based on the encryption seed and N frequency domain face images to obtain the processing result of the vector element, where the processing result is the encryption key or one of the encrypted frequency domain face images.

In some embodiments, the setting order of the column vector to be encrypted may be: the encryption seed is the first row vector element in the column vector to be encrypted, starting from the second row vector element, the N frequency domain face images are arranged in frequency from The order from low to high determines the respective row in the column vector to be encrypted.

The process of obtaining the encryption key includes: obtaining the first convolution results of each of N frequency domain face images, adding the encryption seed and the N first convolution results to obtain the encryption key.

The acquisition process of the encrypted frequency domain face image i' of the frequency domain face image i among N frequency domain face images includes: determining the target row of the frequency domain face image i in the column vector to be encrypted, and obtaining the front of the target row. The second convolution result of a row of processing results of the first vector element that has been convolutionally encrypted, and the second convolution result is multiplied by the frequency domain face image i to obtain the multiplication result. Gets the third convolution result of the processing result for each first vector element preceding the target row. Obtain the first convolution result of the frequency domain face image of the second vector element located behind the target row that has not undergone convolution encryption processing. Add the multiplication result, the first convolution result of the second vector element and the third convolution result of the first vector element to obtain the encrypted frequency domain face image i', where the values of i and i' are 1 to N.

For example, when N frequency domain face images include one low-frequency face image and three high-frequency face images, set the encryption seed to The low-frequency face image is The three high-frequency face images are then the encryption key Frequency domain face image Corresponding encrypted frequency domain face images The calculation process is as follows.

Among them, the sum of φ(·), η(·) and ρ(·) is an arbitrary function, and exp(·) represents the indicator function.

S24: Perform frequency domain inverse processing on the N encrypted frequency domain face images to obtain the encrypted face image.

For an introduction to step S24, please refer to the relevant content records in the above embodiments, and will not be described again here.

In the embodiment of the present disclosure, a convolution encryption process is performed on the encryption seed and N frequency domain face images to obtain the encryption key of the original face image and the encrypted frequency domain face images of each of the N frequency domain face images. In the embodiment of the present disclosure, a column vector to be encrypted is created to perform convolution encryption processing on the encryption seed and N frequency domain face images. Encrypted frequency domain face images of each of the N frequency domain face images can be obtained, and then the encrypted face can be obtained. image, and obtain the encryption key to achieve reversibility of face image encryption.

Figure 3 is a flow chart of a face image encryption method according to an embodiment of the present disclosure. As shown in Figure 3, the method includes steps S31 to S34.

S31: Perform frequency domain processing on the original face image to obtain N frequency domain face images at different frequencies, where N is a positive integer.

S32, obtain the encrypted seed.

S33. Based on the target face encryption network, perform convolution encryption processing on the encryption seed and N frequency domain face images to obtain the encryption key and the encrypted frequency domain face images of each of the N frequency domain face images.

The encryption seed and N frequency domain face images are used as input data and input into the target face encryption network. The target face encryption network performs convolution encryption processing on the input data to obtain the encrypted frequencies of each of the N frequency domain face images. Domain face image.

For example, Figure 4 is a schematic diagram of a target face encryption network. As shown in Figure 4, discrete wavelet transformation (DWT) is performed on the original image I to obtain four frequency domain face images at different frequencies. , which includes a low-frequency face image _IL and three high-frequency face images And randomly sample the Gaussian normal distribution to obtain the encryption seed s. Will s, I _L , As input data, it is input into the target face encryption network. The target face encryption network performs convolution encryption on the input data to obtain the encryption key k and the frequency domain face image _IL , Respective encrypted frequency domain face images right By performing Inverse Discrete Wavelet Transformation (IDWT), the encrypted image I _en can be obtained.

Regarding the specific calculation process of the convolution encryption process, please refer to the relevant introduction in the embodiment of FIG. 2 of this disclosure, and will not be described again here. It should be noted that the module for index calculation is omitted in Figure 4. During the calculation process, s corresponds to correspond correspond correspond correspond And k corresponds to correspond correspond correspond correspond

In some embodiments, the target face encryption network can choose to deeply analyze the DenseNet module and implement it in combination with the Convolutional Block Attention Module (CBAM) module.

S34: Perform frequency domain inverse processing on the N encrypted frequency domain face images to obtain the encrypted face image.

For an introduction to step S34, please refer to the relevant content records in the above embodiments, and will not be described again here.

In the embodiment of the present disclosure, convolution encryption processing is performed based on the target face encryption network, and encrypted frequency domain face images of each of the N frequency domain face images are obtained. In the embodiment of the present disclosure, a target face encryption network is constructed to perform convolution encryption processing on the encryption seed and N frequency domain face images. Encrypted frequency domain face images of each of the N frequency domain face images can be obtained, and then the encrypted face image can be obtained. face image, and obtain the encryption key to achieve reversibility of face image encryption.

Figure 5 is a flow chart of a face image encryption method according to an embodiment of the present disclosure. As shown in Figure 5, the method includes steps S51 to S56.

S51. Obtain a sample face image set and a sample encryption seed set. The sample face image set includes multiple sample face images, and the sample encryption seed set includes multiple sample encryption seeds.

Face detection and alignment are performed on the original image set, and the unified resolution is 256*256 after cropping, which is used as a sample face image set. In some embodiments, the sample face image set can also be divided into a training set, a verification set, and a test set for model training at different stages.

Randomly sample the Gaussian normal distribution to obtain the sample encryption seed set.

S52: Perform frequency domain processing on the sample face image to obtain N sample frequency domain face images at different frequencies.

Regarding the specific implementation of frequency domain processing, please refer to the relevant introduction in each embodiment of the present disclosure, and will not be described again here. It should be noted that during the training process of the same face encryption network, the sample face images need to be processed in the same frequency domain. For example, the sample face images are subjected to wavelet transformation to obtain sample frequencies at four different frequencies. domain image, which includes one low-frequency face image and three high-frequency face images.

S53: Input the sample encryption seed and N sample frequency domain face images into the face encryption network for convolution encryption processing, and obtain encrypted sample frequency domain face images of each of the N sample frequency domain face images.

S54: Perform frequency domain inverse processing on the N encrypted sample frequency domain face images to obtain the encrypted sample face image.

It should be noted that the sample face image undergoes frequency domain processing and frequency domain inverse processing, which is similar to the frequency domain processing and frequency domain inverse processing of the original face image. For the specific implementation of steps S53 to S54, please refer to various steps in this disclosure. The relevant introduction in the embodiment will not be repeated here.

S55, determining the loss function of the face encryption network based on the sample face image and the encrypted sample face image, the sample frequency domain face image and the corresponding encrypted sample frequency domain face image.

In order to meet the needs of the face identity steganography task, it is necessary to constrain the encrypted face image and the original face image to be as similar as possible, ensure that the low-frequency information of the image before and after encryption is consistent, and constrain the changes in the encryption process to occur at the edge as much as possible, that is, not easily detectable by the human eye. area and ensure the authenticity of the generated image, it is necessary to design a loss function and update the model based on the loss value.

In some embodiments, in order to meet the various needs of the face encryption network, multiple loss functions of the face encryption network can be designed, and the multiple loss functions can be weighted to obtain the total loss function of the face encryption network.

S56, adjust the initial face encryption network based on the loss function, and continue to train the adjusted face encryption network based on the next sample face image and sample encryption seed, until the target face encryption network is obtained after the training.

Based on the loss function, the parameters of the constructed face encryption network are adjusted and optimized, and the next sample face image and sample encryption seed are used to continue training the adjusted face encryption network until the end of training conditions are met and the target face encryption is obtained. network.

In some embodiments, when the encryption effect, visual quality and robustness of the encrypted sample face image meet the set requirements, the face encryption network can be considered to have reached the set number of training times and the training end conditions are met. In some embodiments, when the training duration reaches the set value, it can also be considered that the face encryption network has reached the set number of training times and meets the training end conditions.

In some embodiments, the encryption effect is measured by the cosine similarity of facial features before and after encryption, and the visual quality is measured by four indicators: SSIM, PSNR, LPIPS, and MAE. For example, the encryption effect and visual quality experimental results of the encrypted sample image As shown in the table below.

Table 1 Experimental results of encryption effect and visual quality

The lower the cosine similarity between the obtained features is, the better the encryption effect of the constructed face encryption network is. According to the results in Table 1, it can be proven that the face encryption network of the present disclosure can effectively encrypt face images while ensuring visual quality.

In some embodiments, the encrypted sample face image is scaled, compressed, and converted to grayscale, which are common operations when uploading images on social media, and the cosine similarity between the features of the encrypted sample face image and the sample face image is obtained to determine the face. The robustness of the encryption network. For example, the experimental results of the robustness of the encrypted sample image are shown in the table below.

Table 2 Encryption robustness experimental results

Among them, A represents the sample face image, A' represents the encrypted sample face image, and A" represents the deformed encrypted sample face image. The table shows the cosine similarity of the facial features. The lower the cosine similarity between the obtained features, That is to say, it proves that the encryption effect of the constructed face encryption network is better. According to the results in Table 2, it can be found that the encryption obtained by the face encryption network After the sample face image is scaled, compressed, and converted to grayscale, which are common operations when uploading images on social media, it can maintain the feature similarity with the original image, that is, it is robust to these attack methods.

In the embodiment of the present disclosure, a sample face image set and a sample encryption seed set are obtained. The sample face image set includes multiple sample face images, and the sample encryption seed set includes multiple sample encryption seeds. The sample face images are subjected to frequency domain Process, obtain N sample frequency domain face images at different frequencies, input the sample encryption seeds and N sample frequency domain face images into the face encryption network for convolution encryption processing, and obtain N sample frequency domain faces. Encrypted sample frequency domain face images of each image, perform frequency domain inverse processing on the N encrypted sample frequency domain face images, and obtain the encrypted sample face image. Based on the sample face image and the encrypted sample face image, the sample frequency domain face image Face image and corresponding encrypted sample frequency domain face image, determine the loss function of the face encryption network, adjust the initial face encryption network based on the loss function, and continue to adjust the adjusted face based on the next sample face image and sample encryption seed The face encryption network is trained until the end of training to obtain the target face encryption network. In the embodiment of the present disclosure, the face encryption network is trained according to the needs of face identity steganography. In this way, the target face encryption network can be obtained, so that the encryption effect, visual quality and The robustness meets the set requirements.

Figure 6 is a flow chart of a face image encryption method according to an embodiment of the present disclosure. As shown in Figure 6, the method includes steps S61 to S65.

S61: Obtain the reconstruction loss function and the encryption loss function based on the sample face image and the encrypted sample face image.

Reconstruction loss function It is used to constrain the encrypted image and the original face image to be as similar as possible. The specific calculation method is as follows:

Encrypted loss function The process of obtaining the encryption loss function includes: obtaining the first face feature of the sample face image and the second face feature of the encrypted sample face image, and obtaining the encryption loss function based on the first face feature and the second face feature. The specific calculation method is as follows:

S62: Determine the low-frequency loss function and the high-frequency loss function based on the sample frequency domain face image and the corresponding encrypted sample frequency domain face image.

Low frequency loss function Used to ensure that the low-frequency information of the image before and after encryption is consistent, and the high-frequency loss function It is used to constrain changes in the encryption process to occur as much as possible at the edges, that is, areas that are not easily noticeable to the human eye.

The acquisition process of the low-frequency loss function includes: determining the frequency of the sample frequency domain face image, obtaining the first sample frequency domain face image whose frequency is less than the set frequency threshold, and based on the first sample frequency domain face image and the corresponding The encrypted sample frequency domain face image is used to obtain the low-frequency loss function. The specific calculation method is as follows:

The acquisition process of the high-frequency loss function includes: acquiring a second sample frequency domain face image with a frequency greater than or equal to the set frequency threshold, and based on the second sample frequency domain face image and the corresponding encrypted sample frequency domain face image, Obtain the high-frequency loss function. The specific calculation method is as follows:

S63: Perform pixel value normalization processing on the encrypted sample face image, and obtain a pixel loss function based on the normalized pixel value.

Pixel loss function Used to limit the pixel value of the generated image to be between 0-1 to ensure visual quality. The process of obtaining the pixel loss function includes: obtaining the absolute value of the difference between the normalized pixel value of each pixel of the encrypted sample face image and the set value, and determining the maximum absolute value. Obtain the minimum normalized pixel value of the encrypted sample face image, and obtain the pixel loss function based on the maximum absolute value and the minimum normalized pixel value. The specific calculation method is as follows:

S64, perform adversarial recognition on the sample face image and the encrypted sample face image, obtain the first true and false recognition value of the sample face image and the second true and false recognition value of the encrypted sample face image, and based on the first true and false recognition value value and the second true and false identification value to obtain the adversarial loss function.

Adversarial loss function Includes the first adversarial loss function and the second adversarial loss function The adversarial loss function is used to ensure the authenticity of the generated images.

The process of obtaining the first adversarial function includes: obtaining the first adversarial loss function based on the first true and false identification value and the second true and false identification value. The specific calculation method is as follows:

The process of obtaining the second adversarial function includes: obtaining the second adversarial loss function based on the second true and false identification value. The specific calculation method is as follows:

For the first adversarial loss function and the second adversarial loss function Weighting is performed to obtain the adversarial loss function In some embodiments, the weighting coefficients are set to a ₁ and a ₂ , and the sum of a ₁ and a ₂ is 1, then the specific calculation method of the adversarial loss function is as follows:

S65: Weight the reconstruction loss function, encryption loss function, low-frequency loss function, high-frequency loss function, pixel loss function and adversarial loss function to obtain the loss function of the face encryption network.

In some embodiments, a weighted coefficient is assigned to each loss function according to the importance of the requirements corresponding to each loss function, and the reconstruction loss function, encryption loss function, low-frequency loss function, high-frequency loss function, pixel loss function and adversarial loss function are weighted to obtain the loss function of the face encryption network. The specific calculation method is as follows:

In the embodiment of the present disclosure, the reconstruction loss function and the encryption loss function are obtained based on the sample face image and the encrypted sample face image, and the low-frequency loss function and the encryption loss function are determined based on the sample frequency domain face image and the corresponding encrypted sample frequency domain face image. The high-frequency loss function normalizes the pixel value of the encrypted sample face image, and based on the normalized pixel value, obtains the pixel loss function, performs adversarial recognition on the sample face image and the encrypted sample face image, and obtains the sample person The first true and false recognition value of the face image and the second true and false recognition value of the encrypted sample face image are obtained, and based on the first true and false recognition value and the second true and false recognition value, the adversarial loss function is obtained, and the reconstruction loss function and encryption are The loss function, low-frequency loss function, high-frequency loss function, pixel loss function and adversarial loss function are weighted to obtain the loss function of the face encryption network. In the embodiment of the present disclosure, a loss function is designed according to the needs of the face identity steganographic task. The face encryption network is adjusted based on the loss function, which can improve the encryption effect, visual quality and robustness of the face image encrypted through the target face encryption network. The stickiness meets the set requirements.

The above embodiment introduces the encryption method of the face image. Correspondingly, based on the process and calculation formula in the encryption method of the face image, the process and calculation formula in the decryption method of the face image can be deduced to realize the face identity. The reversibility of encryption is implemented in the following examples:

Figure 7 is a flow chart of a method for decrypting a face image according to an embodiment of the present disclosure. As shown in Figure 7, the method includes steps S71 to S74.

S71, perform frequency domain processing on the encrypted face image, and obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer.

Regarding the specific implementation of frequency domain processing, please refer to the relevant introduction in each embodiment of the present disclosure, and will not be described again here.

S72, obtain the encryption key.

When encrypting the original face image, the encryption key of the original face image can be obtained at the same time, which is used to decrypt the encrypted face image. In some embodiments, encryption keys may be obtained from third parties who have permission to use the identity information and from the encryptor themselves.

S73: Perform convolution and decryption processing on the encryption key and N encrypted frequency domain face images to obtain frequency domain face images of each of the N encrypted frequency domain face images.

As a possible implementation method, construct a decryption model of face images, use the obtained encryption key and N encrypted frequency domain face images as network input, perform convolution decryption processing through the model, and obtain N encrypted frequency domain face images. The face images are their respective frequency domain face images.

In some embodiments, a decryption model for face images can also be designed based on the traditional reversible model, and one model can be used to efficiently implement encryption and decryption of face images.

S74: Perform frequency domain inverse processing on the N frequency domain face images to obtain the original face image.

In the previous steps, frequency domain processing was performed on the encrypted face image to restore the image transformation in the high-frequency area. Correspondingly, after the decryption is completed, it is necessary to perform frequency domain inverse processing on the N frequency domain face images to obtain the original face. image.

In this disclosed embodiment, the encrypted face image is subjected to frequency domain processing, and N encrypted frequency domain face images at different frequencies are obtained. N is a positive integer, the encryption key is obtained, and the encryption key and N encrypted frequency domain images are obtained. The face image is subjected to convolution decryption processing to obtain the frequency domain face image of each of the N encrypted frequency domain face images. The N frequency domain face images are subjected to frequency domain inverse processing to obtain the original face image. In the embodiment of the present disclosure, inverse processing is performed on the basis of face image encryption, which can decrypt the encrypted face image and restore the original face image. For the first time, the reversible model is applied to the face identity steganography task to achieve utilization A model efficiently implements encryption and decryption of face images.

Figure 8 is a flow chart of a method for decrypting a face image according to an embodiment of the present disclosure. As shown in Figure 8, the method includes steps S81 to S84.

S81, perform frequency domain processing on the encrypted face image, and obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer.

S82, obtain the encryption key.

For the introduction of steps S81 to S82, please refer to the relevant content records in the above embodiments, and will not be described again here.

S83: Perform convolution and decryption processing on the encryption key and N encrypted frequency domain face images to obtain frequency domain face images and encryption seeds of each of the N encrypted frequency domain face images.

By performing convolution and decryption processing on the encryption key and N encrypted frequency domain face images, not only the frequency domain face images of each of the N encrypted frequency domain face images can be obtained, but also the encryption seeds can be obtained. This encryption seed is used as input at the same time as the original face image when encrypting face identity.

As a possible implementation method, combine the encryption key and N encrypted frequency domain face images in a set order to obtain a column vector to be decrypted, and treat the vectors in the decrypted column vector row by row in reverse order starting from the last row. elements, respectively, perform convolution decryption processing based on the encryption key and N encrypted frequency domain face images to obtain the decryption result of the vector element, where the decryption result is the encryption seed of the encryption key or one of the frequency domain face images. .

In some embodiments, the setting order of the column vector to be decrypted can be: the encryption key is the first row vector element in the column vector to be decrypted, and starting from the second row vector element, the N encrypted frequency domain face images are determined in order from low to high frequency to determine their respective rows in the column vector to be decrypted.

The acquisition process of the frequency domain face image i of the encrypted frequency domain face image i' among the N encrypted frequency domain face images includes: determining the target row of the encrypted frequency domain face image i' in the column vector to be decrypted, and obtaining A first convolution result of the decrypted result of the first vector element that has been subjected to convolution decryption located behind the target row, and a third convolution result of the second vector element that has not been subjected to convolution decryption located in front of the target row. Gets the second convolution result of the second vector element of the row preceding the target row. Subtract the encrypted frequency domain face image i' from the first convolution result of the first vector element and the third convolution result of the second vector element to obtain a subtraction result, and combine the subtraction result with the second convolution result Multiply to obtain the frequency domain face image i of the encrypted frequency domain face image i'.

The process of obtaining the encryption seed includes: subtracting the encryption key from the first convolution results of N frequency domain face images to obtain the encryption seed.

For example, when N encrypted frequency domain face images include one encrypted low-frequency face image and three encrypted high-frequency face images, the encryption key is set to The encrypted low-frequency face image is The three encrypted high-frequency face images are Then encrypt the seed Encrypted frequency domain face image Corresponding frequency domain face images The calculation process is as follows.

S84: Perform frequency domain inverse processing on the N frequency domain face images to obtain the original face image.

For an introduction to step S84, please refer to the relevant content records in the above embodiments, and will not be described again here.

In the embodiment of the present disclosure, a convolution decryption process is performed on the encryption key and N encrypted frequency domain face images to obtain frequency domain face images and encryption seeds for each of the N encrypted frequency domain face images. In the embodiment of the present disclosure, a column vector to be decrypted is created to perform convolution decryption processing on the encryption key and N encrypted frequency domain face images. Frequency domain face images of each of the N encrypted frequency domain face images can be obtained, and the original face image can be restored. face image.

Figure 9 is a flow chart of a method for decrypting a face image according to an embodiment of the present disclosure. As shown in Figure 9, the method includes steps S91 to S94.

S91, perform frequency domain processing on the encrypted face image, and obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer.

S92, obtain the encryption key.

S93, based on the target face decryption network, perform convolution decryption processing on the encryption key and N encrypted frequency domain face images to obtain the encryption seed and the frequency domain face images of the N encrypted frequency domain face images.

The encryption key and N encrypted frequency domain face images are used as input data and input into the target face decryption network. The target face decryption network performs convolution and decryption processing on the input data to obtain each of the N encrypted frequency domain face images. Frequency domain face image.

It should be noted that the target face decryption network and target face encryption network in the embodiments of the present disclosure can be designed based on the existing two-branch reversible model, and one model can be used to efficiently implement encryption and decryption of face images.

For example, Figure 10 is a schematic diagram of a target face decryption network designed based on the above target face encryption network. As shown in Figure 10, the encrypted image I _en is subjected to discrete wavelet transformation (Discrete Wavelet Transformation, DWT), obtain four encrypted frequency domain face images at different frequencies, including an encrypted low-frequency face image and three high-frequency face images And get the encryption key k. Will k, As input data, it is input into the target face decryption network. The target face decryption network performs convolution decryption processing on the input data to obtain the encryption seed s and the encrypted frequency domain face image. Respective frequency domain face images right By performing Inverse Discrete Wavelet Transformation (IDWT), the decrypted image I _de can be obtained.

Regarding the specific calculation process of the convolution decryption process, please refer to the relevant introduction in the embodiment of FIG. 9 of the present disclosure, and will not be described again here. It should be noted that the module for index calculation is omitted in Figure 11. During the calculation process, s corresponds to correspond correspond correspond correspond And k corresponds to correspond correspond correspond correspond

In some embodiments, the target face decryption network can choose to deeply parse the DenseNet module and implement it in combination with the Convolutional Block Attention Module (CBAM) module.

S94: Perform frequency domain inverse processing on the N frequency domain face images to obtain the original face image.

For the introduction of step S94, please refer to the relevant contents in the above embodiment, which will not be repeated here.

In the embodiment of the present disclosure, convolution decryption processing is performed based on the target face decryption network, and frequency domain face images of each of N encrypted frequency domain face images are obtained. In the embodiment of the present disclosure, a target face decryption network is constructed to perform convolution and decryption processing on the encryption key and N encrypted frequency domain face images. Frequency domain face images of each of the N encrypted frequency domain face images can be obtained, and then restored Original face image.

FIG. 11 is a flowchart of a method for decrypting a facial image according to an embodiment of the present disclosure. As shown in FIG. 11 , the method includes steps S111 to S116 .

S111. Obtain a sample encrypted face image set and a sample encryption key set. The sample face image set includes multiple sample encrypted face images, and the sample encryption key set includes multiple sample encryption keys.

S112: Perform frequency domain processing on the sample encrypted face image to obtain N sample encrypted frequency domain face images at different frequencies.

S113, input the sample encryption key and N sample encrypted frequency domain face images into the face decryption network for convolution decryption processing, and obtain sample frequency domain face images of each of the N sample encrypted frequency domain face images.

S114: Perform frequency domain inverse processing on N sample frequency domain face images to obtain sample face images.

S115. Determine the loss function of the face decryption network based on the sample encrypted face image and the sample face image, the sample encrypted frequency domain face image and the corresponding sample frequency domain face image.

Regarding the specific implementation of steps S112 to S115, please refer to the relevant introductions in the embodiments of the present disclosure, and will not be described again here.

S116, adjust the initial face decryption network based on the loss function, and continue to train the adjusted face decryption network based on the next sample encrypted face image and sample encryption key until the end of the training to obtain the target face decryption network.

Adjust and optimize the parameters of the constructed face decryption network based on the loss function, and return to use the next encrypted face image and sample encryption key to continue training the adjusted face decryption network until the end of training conditions are met to obtain the target face Decrypt the network.

In some embodiments, when the decryption effect and visual quality of the sample face image meet the set requirements, the face decryption network can be considered to have reached the set number of training times and meet the training end conditions. In some embodiments, when the training duration reaches the set value, it can also be considered that the face decryption network has reached the set number of training times and the training end conditions are met.

In some embodiments, the decryption effect is measured by the cosine similarity of facial features between the decrypted image and the original image, and the visual quality is measured by three indicators: SSIM, LPIPS, and MAE. For example, the decryption effect of the sample image and the visual quality experimental results As shown in the table below.

Table 3 Recovery quality of decrypted images

The higher the cosine similarity between the obtained features is, the better the decryption effect of the constructed face decryption network is. According to the results in Table 3, it can be proven that the face decryption network of the present disclosure can effectively decrypt face images while ensuring visual quality.

At the same time, if a mismatched encryption key is used, the original image cannot be correctly restored. The uniqueness and non-replicability of the key ensures the security of encryption.

In this disclosed embodiment, a sample encrypted face image set and a sample encryption key set are obtained, the sample face image set includes multiple sample encrypted face images, the sample encryption key set includes multiple sample encryption keys, and the sample is encrypted The face image is processed in the frequency domain, and N sample encrypted frequency domain face images at different frequencies are obtained. The sample encryption key and N sample encrypted frequency domain face images are input into the face decryption network for convolution decryption processing. , obtain each sample frequency domain face image of N sample encrypted frequency domain face images, perform frequency domain inverse processing on the N sample frequency domain face images, and obtain the sample face image, based on the sample encrypted face image and sample face Face image, sample encrypted frequency domain face image and corresponding sample frequency domain face image, determine the loss function of the face decryption network, adjust the initial face decryption network based on the loss function, and encrypt the face image and sample based on the next sample The encryption key continues to train the adjusted face decryption network until the target face decryption network is obtained after training. In the embodiment of the present disclosure, the face decryption network is trained according to the requirement of reversible face encryption. In this way, the target face decryption network can be obtained, so that the decryption effect and visual quality of the face image decrypted by the target face decryption network meet the requirements of the design. Set requirements.

Figure 12 is a structural diagram of a face image encryption device according to an embodiment of the present disclosure. As shown in Figure 12, the face image encryption device 120 includes a frequency domain processing module 121, an acquisition module 122, an encryption module 123 and a frequency domain Inverse processing module 124.

The frequency domain processing module 121 is used to perform frequency domain processing on the original face image and obtain N frequency domain face images at different frequencies, where N is a positive integer.

Obtaining module 122 is used to obtain encrypted seeds.

The encryption module 123 is used to perform convolution encryption processing on the encryption seed and N frequency domain face images to obtain encrypted frequency domain face images of each of the N frequency domain face images.

The frequency domain inverse processing module 124 is used to perform frequency domain inverse processing on N encrypted frequency domain face images to obtain encrypted face images.

It should be noted that the foregoing explanation of the embodiment of the facial image encryption method is also applicable to the facial image encryption device of this embodiment, and will not be described again here.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the encryption module 123 is also used to perform convolution encryption processing on the encryption seed and N frequency domain face images to obtain the original face image. Encryption key.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the encryption module 123 is also used to: combine the encryption seed and N frequency domain face images in a set order to obtain a column to be encrypted Vector; starting from the first row, the vector elements in the encrypted column vector are treated row by row in positive order, and convolution encryption is performed based on the encryption seed and N frequency domain face images to obtain the processing result of the vector element, where the processing result is the encryption key or is one of the encrypted frequency domain face images.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the setting order of the column vector to be encrypted is: the encryption seed is the first row vector element in the column vector to be encrypted, starting from the second row vector element , N frequency domain face images determine their respective rows in the column vector to be encrypted in order from low to high frequency.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the encryption module 123 is also used to: obtain the first convolution results of each of the N frequency domain face images; The convolution results are added to obtain the encryption key.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the encryption module 123 is also used to: determine the target row of the frequency domain face image i in the column vector to be encrypted; obtain the previous row of the target row The second convolution result of the processing result of the first vector element that has undergone convolution encryption processing, and the second convolution result is multiplied with the frequency domain face image i to obtain the multiplication result; obtain each of the first vector elements located in front of the target row The third convolution result of the processing result of a vector element; obtain The first convolution result of the frequency domain face image of the second vector element located behind the target row that has not undergone convolution encryption processing; the multiplication result, the first convolution result of the second vector element and the first convolution result of the first vector element The three convolution results are added to obtain the encrypted frequency domain face image i', where the values of i and i' range from 1 to N.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the encryption module 123 is also used to: input the encryption seed and N frequency domain face images as input data into the target face encryption network, The target face encryption network performs convolution encryption on the input data to obtain the encrypted frequency domain face images of each of the N frequency domain face images.

In some embodiments, in a possible implementation of the disclosed embodiment, as shown in FIG. 13 , the face image encryption device 120 further includes: a training module 125, which is used to: obtain a sample face image set and a sample encryption seed set, wherein the sample face image set includes a plurality of sample face images, and the sample encryption seed set includes a plurality of sample encryption seeds; perform frequency domain processing on the sample face images to obtain N sample frequency domain face images at different frequencies; input the sample encryption seeds and the N sample frequency domain face images into a face encryption network for convolution encryption processing to obtain encrypted sample frequency domain face images of the N sample frequency domain face images; perform frequency domain inverse processing on the N encrypted sample frequency domain face images to obtain encrypted sample face images; determine a loss function of the face encryption network based on the sample face images and the encrypted sample face images, the sample frequency domain face images and the corresponding encrypted sample frequency domain face images; adjust the initial face encryption network based on the loss function, and continue to train the adjusted face encryption network based on the next sample face image and the sample encryption seed until the training is completed to obtain the target face encryption network.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the training module 125 is also used to: obtain the reconstruction loss function and the encryption loss function according to the sample face image and the encrypted sample face image; Frequency domain face image and corresponding encrypted sample frequency domain face image, determine the low-frequency loss function and high-frequency loss function; perform pixel value normalization processing on the encrypted sample face image, and obtain pixels based on the normalized pixel value Loss function; perform adversarial recognition on the sample face image and the encrypted sample face image, obtain the first true and false recognition value of the sample face image and the second true and false recognition value of the encrypted sample face image, and based on the first true and false The recognition value and the second true and false recognition value are used to obtain the adversarial loss function; the reconstruction loss function, encryption loss function, low-frequency loss function, high-frequency loss function, pixel loss function and adversarial loss function are weighted to obtain the loss of the face encryption network function.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the training module 125 is also used to: obtain the first facial feature of the sample face image and the second facial feature of the encrypted sample face image. , and obtain the encryption loss function based on the first face feature and the second face feature.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the training module 125 is also used to: determine the frequency of the face image in the sample frequency domain; obtain the first sample frequency whose frequency is less than the set frequency threshold. domain face image, and obtain the low-frequency loss function based on the first sample frequency domain face image and the corresponding encrypted sample frequency domain face image; obtain the second sample frequency domain face image with a frequency greater than or equal to the set frequency threshold , and obtain the high-frequency loss function based on the second sample frequency domain face image and the corresponding encrypted sample frequency domain face image.

In some embodiments, in a possible implementation method of the disclosed embodiment, the training module 125 is also used to: obtain the absolute value of the difference between the normalized pixel value of each pixel of the encrypted sample face image and the set value, and determine the maximum absolute value; obtain the minimum normalized pixel value of the encrypted sample face image; and obtain the pixel loss function based on the maximum absolute value and the minimum normalized pixel value.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the training module 125 is also used to: obtain the first adversarial loss function based on the first true and false identification value and the second true and false identification value; The second true and false identification value is used to obtain the second adversarial loss function; the first adversarial loss function and the second adversarial loss function are weighted to obtain the adversarial loss function.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the frequency domain processing module 121 is also used to: perform frequency domain decomposition on the original face image based on wavelet transform to obtain N frequency domain faces. image.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the acquisition module 122 is also configured to randomly sample the Gaussian normal distribution to obtain encryption seeds.

Figure 14 is a structural diagram of a face image decryption device according to an embodiment of the present disclosure. As shown in Figure 14, the face image decryption device 130 includes a frequency domain processing module 131, an acquisition module 132, a decryption module 133 and a frequency domain Inverse processing module 134.

The frequency domain processing module 131 is used to perform frequency domain processing on the encrypted face image and obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer.

Obtaining module 132 is used to obtain the encryption key.

The decryption module 133 is used to perform convolution and decryption processing on the encryption key and N encrypted frequency domain face images to obtain frequency domain face images of each of the N encrypted frequency domain face images.

The frequency domain inverse processing module 134 is used to perform frequency domain inverse processing on N frequency domain face images to obtain original face images.

It should be noted that the foregoing explanation of the embodiment of the facial image decryption method is also applicable to the facial image decryption device of this embodiment, and will not be described again here.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the decryption module 133 is also configured to: combine the encryption key and N encrypted frequency domain face images in a set order to obtain a to-be-received face image. Decrypt the column vector; treat the vector elements in the decrypted column vector row by row in reverse order starting from the last row, and perform convolution decryption processing based on the encryption key and N encrypted frequency domain face images to obtain the decryption result of the vector elements, where , the decryption result is the encryption seed of the encryption key or one of the frequency domain face images.

In some embodiments, in a possible implementation method of the embodiments of the present disclosure, the setting order of the column vector to be decrypted is: the encryption key is the first row vector element in the column vector to be decrypted, and starting from the second row vector element, the N encrypted frequency domain face images are ordered from low to high in frequency to determine their respective rows in the column vector to be decrypted.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the decryption module 133 is also used to: determine the target row of the encrypted frequency domain face image i' in the column vector to be decrypted; obtain the target row located in The first convolution result of the decryption result of the subsequent first vector element that has undergone convolution decryption processing, and the third convolution result of the second vector element that has not undergone convolution decryption processing in front of the target row; obtain the target row's The second convolution result of the second vector element in the previous row; subtract the encrypted frequency domain face image i' from the first convolution result of the first vector element and the third convolution result of the second vector element to obtain Subtract the result, and multiply the subtraction result with the second convolution result to obtain the frequency domain face image i of the encrypted frequency domain face image i'.

In some embodiments, in a possible implementation of this embodiment of the present disclosure, the decryption module 133 is also configured to: subtract the encryption key from the first convolution results of the N frequency domain face images to obtain the encrypted seed.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the decryption module 133 is also used to: input the encryption key and N encrypted frequency domain face images as input data into the target face decryption network. In , the target face decryption network performs convolution decryption processing on the input data to obtain the frequency domain face images of each of the N encrypted frequency domain face images.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, as shown in Figure 15, the face image decryption device 130 also includes: a training module 135 for: obtaining a sample encrypted face image set and Sample encryption key set. The sample face image set includes multiple sample encrypted face images. The sample encryption key set includes multiple sample encryption keys. Frequency domain processing is performed on the sample encrypted face image to obtain N different frequencies. The sample encrypted frequency domain face image; the sample encryption key and N sample encrypted frequency domain face images are input into the face decryption network for convolution decryption processing, and each sample of the N sample encrypted frequency domain face image is obtained. Frequency domain face images; perform frequency domain inverse processing on N sample frequency domain face images to obtain sample face images; encrypt face images and sample face images based on samples, sample encrypted frequency domain face images and corresponding samples Frequency domain face image, determine the loss function of the face decryption network; adjust the initial face decryption network based on the loss function, and continue to train the adjusted face decryption network based on the next sample encrypted face image and sample encryption key , until the end of training to obtain the target face decryption network.

In some embodiments, in a possible implementation of the embodiment of the present disclosure, the training module 135 is also used to: obtain the reconstruction loss function and the encryption loss function according to the sample face image and the encrypted sample face image; Frequency domain face image and corresponding encrypted sample frequency domain face image, determine the low-frequency loss function and high-frequency loss function; perform pixel value normalization processing on the encrypted sample face image, and obtain pixels based on the normalized pixel value Loss function; perform adversarial recognition on the encrypted sample face image, obtain the first true and false recognition value of the sample face image and the second true and false recognition value of the encrypted sample face image, and based on the first true and false recognition value and the second true and false recognition value Two true and false recognition values, obtain the adversarial loss function; The construction loss function, encryption loss function, low-frequency loss function, high-frequency loss function, pixel loss function and adversarial loss function are weighted to obtain the loss function of the face encryption network.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, a computer program product, and a computer program.

According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, including a memory and a processor; wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, to Encryption/decryption method for facial images used to implement any embodiment of the present disclosure.

16 illustrates a schematic block diagram of an example electronic device 140 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to refer to various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are examples only and are not intended to limit implementations of the disclosure described and/or claimed herein.

As shown in FIG. 16 , the system includes a memory 141 , a processor 142 , and a computer program stored in the memory 141 and executable on the processor 142 . When the processor 142 executes the program, the aforementioned facial image encryption/decryption method is implemented.

According to an embodiment of the present disclosure, the present disclosure also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the face image encryption/decryption method of any embodiment of the present disclosure is implemented. .

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

To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or a trackball) through which a user can provide input to the computer. Other kinds of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and may be provided in any form, including Acoustic input, voice input or tactile input) to receive input from the user.

The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., A user's computer having a graphical user interface or web browser through which the user can interact with implementations of the systems and technologies described herein), or including such backend components, middleware components, or any combination of front-end components in a computing system. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communications network). Examples of communication networks include: local area network (LAN), wide area network (WAN), and the Internet.

Computer systems may include clients and servers. Clients and servers are generally remote from each other and typically interact over a communications network. The relationship of client and server is created by computer programs running on corresponding computers and having a client-server relationship with each other. The server can be a cloud server, a distributed system server, or a server combined with a blockchain.

According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, including a computer program. When executed by a processor, the computer program implements the encryption/decryption method of a face image as in any embodiment of the present disclosure.

According to an embodiment of the present disclosure, the present disclosure also provides a computer program. The computer program includes a computer program code. When the computer program code is run on a computer, it causes the computer to perform the facial image processing according to any embodiment of the present disclosure. Encryption/decryption methods.

It should be noted that the foregoing explanations of the method and device embodiments also apply to the electronic equipment, computer-readable storage media, computer program products and computer programs of the above-mentioned embodiments, and will not be described again here.

In addition, the terms “first” and “second” are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of indicated technical features. Therefore, features defined as "first" and "second" may explicitly or implicitly include one or more of these features. In the description of the present invention, "plurality" means two or more than two, unless otherwise explicitly and specifically limited.

In the description of this specification, reference to the terms "one embodiment," "some embodiments," "an example," "specific examples," or "some examples" or the like means that specific features are described in connection with the embodiment or example. , structures, materials or features are included in at least one embodiment or example of the invention. In this specification, the schematic expressions of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, those skilled in the art may combine and combine different embodiments or examples and features of different embodiments or examples described in this specification unless they are inconsistent with each other.

Although the embodiments of the present invention have been shown and described above, it can be understood that the above-mentioned embodiments are illustrative and should not be construed as limitations of the present invention. Those of ordinary skill in the art can make modifications to the above-mentioned embodiments within the scope of the present invention. The embodiments are subject to changes, modifications, substitutions and variations.

All embodiments of the present disclosure can be executed alone or in combination with other embodiments, which are considered to be within the scope of protection claimed by the present disclosure.

Claims

A facial image encryption method, which is characterized by including:

Perform frequency domain processing on the original face image to obtain N frequency domain face images at different frequencies, where N is a positive integer;

Get encrypted seeds;

Perform convolution encryption processing on the encryption seed and the N frequency domain face images to obtain encrypted frequency domain face images of each of the N frequency domain face images;

Perform frequency domain inverse processing on the N encrypted frequency domain face images to obtain encrypted face images.
The method of claim 1, further comprising:

Perform convolution encryption processing on the encryption seed and the N frequency domain face images to obtain the encryption key of the original face image.
The method according to claim 1 or 2, characterized in that the process of performing convolution encryption processing on the encryption seed and N frequency domain face images includes:

Combine the encryption seed and the N frequency domain face images in a set order to obtain a column vector to be encrypted;

Starting from the first row, perform convolution encryption processing on the vector elements in the column vector to be encrypted row by row in positive order based on the encryption seed and the N frequency domain face images to obtain the vector elements. Processing result, wherein the processing result is the encryption key or one of the encrypted frequency domain face images.
The method according to claim 2 or 3, characterized in that the process of obtaining the encryption key comprises:

Obtain the first convolution results of each of the N frequency domain face images;

The encryption seed and N first convolution results are added to obtain the encryption key.
The method according to claim 3 or 4, characterized in that the acquisition process of the encrypted frequency domain face image i' of the frequency domain face image i among the N frequency domain face images includes:

Determine the target row of the frequency domain face image i in the column vector to be encrypted;

Obtain the second convolution result of the processing result of the first vector element of the previous row of the target row that has undergone the convolution encryption process, and combine the second convolution result with the frequency domain face image i Multiply to get the multiplication result;

Obtaining a third convolution result of the processing result of each first vector element located in front of the target row;

Obtain the first convolution result of the frequency domain face image of the second vector element located behind the target row that has not undergone the convolution encryption process;

The multiplication result, the first convolution result of the second vector element and the third convolution result of the first vector element are added to obtain the encrypted frequency domain face image i', where The values of i and i' are from 1 to N.
The method according to claim 5, characterized in that: performing convolution encryption processing on the encryption seed and the N frequency domain face images to obtain the encrypted frequency of each of the N frequency domain face images. Domain face images, including:

The encryption seed and N frequency domain face images are used as input data and input into the target face encryption network. The target face encryption network performs convolution encryption processing on the input data to obtain N all the face images. The respective encrypted frequency domain face images of the frequency domain face images are described.
The method according to claim 6, characterized in that the training process of the target face encryption network includes:

Obtain a sample face image set and a sample encryption seed set, the sample face image set includes a plurality of sample face images, and the sample encryption seed set includes a plurality of sample encryption seeds;

Perform frequency domain processing on the sample face image to obtain N sample frequency domain face images at different frequencies;

The sample encryption seed and the N sample frequency domain face images are input into the face encryption network for convolution encryption processing to obtain the encrypted sample frequency domain face images of each of the N sample frequency domain face images. ;

Perform frequency domain inverse processing on N encrypted sample frequency domain face images to obtain encrypted sample face images;

Determine the loss function of the face encryption network based on the sample face image and the encrypted sample face image, the sample frequency domain face image and the corresponding encrypted sample frequency domain face image;

The initial face encryption network is adjusted based on the loss function, and the adjusted face encryption network is continued to be trained based on the next sample face image and sample encryption seed, until the target face encryption network is obtained after training is completed.
The method according to claim 7, characterized in that, based on the sample face image and the encrypted sample face image, the sample frequency domain face image and the corresponding encrypted sample frequency domain face image, Determine the loss function of the face encryption network, including:

Obtain a reconstruction loss function and an encryption loss function according to the sample face image and the encrypted sample face image;

Determine the low-frequency loss function and the high-frequency loss function according to the sample frequency domain face image and the corresponding encrypted sample frequency domain face image;

Perform pixel value normalization processing on the encrypted sample face image, and obtain a pixel loss function based on the normalized pixel value;

Perform adversarial recognition on the sample face image and the encrypted sample face image, obtain the first true and false recognition value of the sample face image and the second true and false recognition value of the encrypted sample face image, and Based on the first true and false identification value and the second true and false identification value, obtain an adversarial loss function;

The reconstruction loss function, the encryption loss function, the low-frequency loss function, the high-frequency loss function, the pixel loss function and the adversarial loss function are weighted to obtain the loss function of the face encryption network .
The method according to any one of claims 1 to 8, characterized in that performing frequency domain processing on the original face image to obtain N frequency domain face images at different frequencies includes:

The original face image is decomposed in frequency domain based on wavelet transform to obtain N frequency domain face images.
A method for decrypting face images, which is characterized by including:

Perform frequency domain processing on the encrypted face image to obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer;

Get the encryption key;

Perform convolution and decryption processing on the encryption key and the N encrypted frequency domain face images to obtain frequency domain face images of each of the N encrypted frequency domain face images;

Perform frequency domain inverse processing on the N frequency domain face images to obtain original face images.
The method according to claim 10, characterized in that the process of performing convolution decryption processing on the encryption key and the N encrypted frequency domain face images includes:

The encryption key and the N encrypted frequency domain face images are combined according to the set order to obtain a column vector to be decrypted;

Starting from the last row, perform convolution decryption processing on the vector elements in the column vector to be decrypted row by row in reverse order based on the encryption key and the N encrypted frequency domain face images to obtain the vector elements. The decryption result, wherein the decryption result is the encryption seed of the encryption key or one of the frequency domain face images.
The method according to claim 10 or 11, characterized in that: performing convolution and decryption processing on the encryption key and N encrypted frequency domain face images to obtain N encrypted frequency domain face images. The respective frequency domain face images of the images, including:

The encryption key and the N encrypted frequency domain face images are used as input data and input into the target face decryption network. The target face decryption network performs convolution decryption processing on the input data to obtain N Frequency domain face images of each of the encrypted frequency domain face images.
The method according to claim 12, characterized in that the training process of the target face decryption network includes:

Acquire a sample encrypted face image set and a sample encryption key set, wherein the sample face image set includes a plurality of sample encrypted face images, and the sample encryption key set includes a plurality of sample encryption keys;

Perform frequency domain processing on the sample encrypted face image to obtain N sample encrypted frequency domain face images at different frequencies;

Input the sample encryption key and the N sample encrypted frequency domain face images into a face decryption network for convolution decryption processing to obtain sample frequency domain face images of each of the N sample encrypted frequency domain face images;

Perform frequency domain inverse processing on N sample frequency domain face images to obtain sample face images;

Determine a loss function of the face decryption network based on the sample encrypted face image and the sample face image, the sample encrypted frequency domain face image and the corresponding sample frequency domain face image;

Adjust the initial face decryption network based on the loss function, and continue to train the adjusted face decryption network based on the next sample encrypted face image and sample encryption key until the end of training to obtain the target face decryption network.
The method according to claim 13, characterized in that the encrypted face image based on the sample and the sample face image, the sample encrypted frequency domain face image and the corresponding sample frequency domain face image, Determine the loss function of the face decryption network, including:

Obtain a reconstruction loss function and an encryption loss function according to the sample face image and the encrypted sample face image;

Determine the low-frequency loss function and the high-frequency loss function according to the sample frequency domain face image and the corresponding encrypted sample frequency domain face image;

Perform pixel value normalization processing on the encrypted sample face image, and obtain a pixel loss function based on the normalized pixel value;

Perform adversarial recognition on the encrypted sample face image, obtain the first true and false recognition value of the sample face image and the second true and false recognition value of the encrypted sample face image, and based on the third A true and false identification value and the second true and false identification value to obtain an adversarial loss function;

The reconstruction loss function, the encryption loss function, the low-frequency loss function, the high-frequency loss function, the pixel loss function and the adversarial loss function are weighted to obtain the loss function of the face encryption network .
A face image encryption device, characterized by including:

A frequency domain processing module, used to perform frequency domain processing on the original face image and obtain N frequency domain face images at different frequencies, where N is a positive integer;

An acquisition module, used to obtain an encrypted seed;

An encryption module, configured to perform convolution encryption processing on the encryption seed and the N frequency domain face images, to obtain encrypted frequency domain face images of each of the N frequency domain face images;

A frequency domain inverse processing module is used to perform frequency domain inverse processing on the N encrypted frequency domain face images to obtain an encrypted face image.
A device for decrypting facial images, which is characterized by including:

A frequency domain processing module, used to perform frequency domain processing on the encrypted face image, and obtain N encrypted frequency domain face images at different frequencies, where N is a positive integer;

Obtain module, used to obtain encryption keys;

A decryption module, configured to perform convolution and decryption processing on the encryption key and the N encrypted frequency domain face images, to obtain frequency domain face images of each of the N encrypted frequency domain face images;

A frequency domain inverse processing module is used to perform frequency domain inverse processing on the N frequency domain face images to obtain original face images.
An electronic device, characterized by including a memory and a processor;

The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the method according to any one of claims 1 to 9 or claims 10 to 14.
A computer-readable storage medium with a computer program stored thereon, characterized in that when the program is executed by a processor, the method as described in any one of claims 1 to 9 or 10 to 14 is implemented.
A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 9 or 10 to 14.
A computer program, characterized in that the computer program includes computer program code, and when the computer program code is run on a computer, the computer executes any one of claims 1 to 9 or 10 to 14 the method described.