A Chaotic Image Selective Encryption Method Based on Facial Recognition and Key Optimization

Abstract

This invention belongs to the field of digital image encryption technology and discloses a selective encryption method for chaotic images based on facial recognition and key optimization. The method includes the following steps: Based on an existing chaotic mapping system model, an exponential term and a modulo function are added to optimize the system model's performance, establishing a three-dimensional amplitude-limited coupled hyperchaotic mapping system model; based on the three-dimensional amplitude-limited coupled hyperchaotic mapping system model, a color image face encryption scheme is designed. This invention employs the aforementioned selective encryption method for chaotic images based on facial recognition and key optimization to establish a three-dimensional amplitude-limited coupled hyperchaotic mapping system model, selectively encrypting facial regions, resulting in a wide range of chaos and high complexity; it adopts a face recognition encryption scheme based on the DeepFace model to effectively improve encryption speed; and it uses the PSO optimization algorithm to effectively resist known plaintext attacks, offering significant advantages in computational efficiency and encryption quality compared to traditional encryption methods.

Description

Technical Field

[0001] This invention relates to the field of digital image encryption technology, and in particular to a method for selective encryption of chaotic images based on facial recognition and key optimization. Background Technology

[0002] With the continuous advancement of computer technology and artificial intelligence, the amount of information on the internet has experienced explosive growth. Massive amounts of data, including images, videos, files, and chat logs, need to be transmitted through public channels, making the leakage of personal privacy a major challenge for everyone. Therefore, secure communication has become a hot research topic in modern society. Image encryption, with its high redundancy, large data capacity, and strong information correlation, places more stringent requirements on encryption schemes. Therefore, many scholars are dedicated to finding encryption schemes that are more efficient and less vulnerable to attack than traditional solutions. Technologies such as DNA encoding, compressed sensing, and information hiding have been developed and widely combined for image encryption.

[0003] Since the concept of chaos was first proposed in 1963, chaos theory has developed rapidly, gradually evolving into a mature and practical theoretical framework. Chaotic systems are unpredictable and sensitive to initial conditions. This means that the long-term iterative results of a chaotic system are unpredictable, and small changes in the initial state can lead to huge changes in its dynamic behavior. These characteristics have made them widely used in secure communication, especially in image encryption. In recent years, many image encryption schemes based on chaos theory have been proposed.

[0004] However, many existing chaotic image encryption schemes have some drawbacks. For example, some chaotic systems lack sufficient chaotic properties, have simple structures, and small parameter ranges, which may lead to chaotic degradation. These problems result in decreased security. Furthermore, traditional encryption methods involve two steps: scrambling and diffusion. Diffusion typically involves an XOR operation between the chaotic sequence and image pixel values. This pixel obfuscation scheme is widely used, making it vulnerable to targeted attacks.

[0005] Most traditional encryption schemes suffer from high time complexity even with good encryption results, making selective encryption particularly important. Facial images carry unique biometric information, making their confidentiality extremely crucial. In applications such as facial recognition and facial unlocking, facial images often contain a large amount of redundant information. Therefore, current research focuses on selectively encrypting facial regions. Summary of the Invention

[0006] The purpose of this invention is to provide a selective encryption method for chaotic images based on facial recognition and key optimization. It establishes a three-dimensional amplitude-limited coupled hyperchaotic mapping system model, selectively encrypts the facial region, and solves the defects of previous proposed chaotic system parameter ranges, excessively long encryption time, and insufficient randomness of the generated sequence.

[0007] To achieve the above objectives, this invention provides a method for selective encryption of chaotic images based on facial recognition and key optimization, comprising the following steps:

[0008] Step S1: Based on the existing chaotic mapping system model, add an exponential term and a modulo function to optimize the performance of the system model and establish a three-dimensional limited-amplitude coupled hyperchaotic mapping system model.

[0009] Step S2: Based on the three-dimensional amplitude-limited coupled hyperchaotic mapping system model, design a color image face encryption scheme.

[0010] Preferably, in step S1, based on the existing chaotic mapping system model, an exponential term and a modulo function are added to optimize the system model's performance, establishing a three-dimensional amplitude-limited coupled hyperchaotic mapping system model. The specific process is as follows:

[0011] Step S11: Based on the existing chaotic mapping system model, add an exponential term and a modulo function to optimize the performance of the system model; among them, the addition of the exponential term and the modulo function plays a role in limiting the amplitude.

[0012] Based on this, a three-dimensional limited-amplitude coupled hyperchaotic mapping system model is established, as shown below:

[0013]

[0014] Where, x n y n z n e is the system's state variable; g These are the control parameters for system complexity; a, b, c, and K are conventional parameters that have little impact on system complexity. As an intermediate variable;

[0015] Step S12: Based on the established three-dimensional limited-amplitude coupled hyperchaotic mapping system model, perform dynamic behavior analysis, specifically including Lyapunov exponential diagram, bifurcation diagram, phase diagram, sample entropy, and NIST test.

[0016] Preferably, in step S2, a color image face encryption scheme is designed based on the three-dimensional amplitude-limited coupled hyperchaotic mapping system model, and the specific process is as follows:

[0017] Step 21: Read the original image and extract a facial image of size M×N×3 based on the previous facial recognition results, and rearrange it to size M×3×N;

[0018] Step 22: Match the identified face image with the images in the database. If the match is successful, continue with the encryption operation.

[0019] Step 23: Use the particle swarm optimization algorithm to search for the optimal system key and control the three-dimensional limited-amplitude coupled hyperchaotic mapping system model to generate the required pseudo-random sequence;

[0020] Step 24: Perform cyclic shifting pixel rearrangement and pixel value obfuscation using half-tensor product operation on the rearranged face image, and then rearrange it to an M×N×3 size.

[0021] Step 25: Embed the encrypted face image back into the original image to obtain the final encrypted result.

[0022] Preferably, in step S22, the FaceNet++ model is used to perform face recognition, comparison, and region extraction.

[0023] First, a face database is built using selected images, and all input images to be encrypted are compared with the images in the database.

[0024] Then, in the FaceNet++ model, the Euclidean distance between two images is compared with a preset threshold to determine whether they belong to the same person.

[0025] A match is considered successful when the calculated maximum Euclidean distance between the input image and all face images in the database is greater than a preset threshold.

[0026] Preferably, in step S23, the particle swarm optimization algorithm is used to search for the optimal system key, and the three-dimensional limited-amplitude coupled hyperchaotic mapping system model is controlled to generate the required pseudo-random sequence. The specific process is as follows:

[0027] Step 231: Use the particle swarm optimization algorithm to search for the optimal system key;

[0028] During the execution of the particle swarm optimization algorithm, the velocity and position of all particles in each generation are updated according to the following formula:

[0029]

[0030] in, Let be the particle position at time t; The position of the particle at time t+1; Let be the particle velocity vector at time t; Let p be the particle velocity vector at time t; pbest is the particle's historical best position; best i The historical best position of the population; w is the inertial weight; c1 and c2 are the parameters that control the particle velocity; r1 and r2 are random numbers before [0,1].

[0031] The particle dimension is set to 8 dimensions, corresponding to the 8 keys required; a 64×64×3 pixel block is randomly extracted from the image and encrypted using the keys. The peak signal-to-noise ratio and Shannon entropy of the encrypted image are calculated using formulas (4) and (5), as shown below:

[0032]

[0033] Where MSE represents the mean square error of the image; x i Input image; The input mean is n; n is the number of pixels; PSNR is the peak signal-to-noise ratio.

[0034] Next, the fitness function of the particle swarm optimization algorithm is defined as follows:

[0035]

[0036] Where F(p) is the fitness; ω1 and ω2 are control coefficients, which are 0.3 and 0.7 respectively;

[0037] The quality of encryption is judged by the size of this fitness. After multiple rounds of iteration, an optimal solution is returned, which is the required system key.

[0038] Step 232: Control the three-dimensional amplitude-limited coupled hyperchaotic mapping system model to generate the required pseudo-random sequence;

[0039] Based on the system key mentioned above, generate three corresponding chaotic sequences x, y, and z; discard the first 500 states of the sequences, and use formula (7) to generate the three subsequent sequences line, row, and val:

[0040]

[0041] Among them, line i row i val i Given three distinct pseudo-random sequences; floor represents the floor function; x i y i z i is the original sequence obtained by iterating through the chaotic system; n and m are the length and width of the image to be encrypted.

[0042] Preferably, in step S24, the rearranged face image undergoes cyclic shifting pixel rearrangement and pixel value obfuscation using half-tensor product operation, and then it is rearranged to a size of M×N×3. The specific process is as follows:

[0043] Step 241: Scramble the pixels of the rearranged face image by alternating row and column shifts;

[0044] Step 242: Pixel value obfuscation based on half-tensor product operation;

[0045] Given two matrices A∈M m×n ,B∈M p×q If n and p have a least common multiple of t, then T is defined as the semi-tensor product of the two matrices, as shown below:

[0046]

[0047] Where M represents the set of all matrices of size m×n; ∝ represents the semi-tensor product operation; represents the tensor product of matrices; I represents the identity matrix.

[0048] Therefore, the present invention employs the above-mentioned chaotic image selective encryption method based on facial recognition and key optimization, and the beneficial effects are as follows:

[0049] (1) The present invention establishes a three-dimensional limited amplitude coupled hyperchaotic mapping system model. The system model has a controllable Lyapunov exponent, a wide range of chaos, high complexity, and selective encryption of facial regions.

[0050] (2) The present invention adopts a face recognition encryption scheme based on the DeepFace model to effectively improve encryption speed;

[0051] (3) The present invention uses the PSO optimization algorithm to control the generation of initial values ​​and control parameters in the system, and generates a key associated with the image to be encrypted, so that the encryption scheme can effectively resist known plaintext attacks.

[0052] (4) In this invention, STP is a novel solution to solve pixel confusion in images. It is very different from the traditional XOR-based confusion method, making it effective against traditional attack methods. Compared with traditional encryption, it has advantages in computational efficiency and encryption quality.

[0053] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0054] Figure 1 This is a flowchart of the overall process of a chaotic image selective encryption method based on facial recognition and key optimization according to the present invention.

[0055] Figure 2 This is a graph showing the Lyapunov exponential variation of different parameters of the three-dimensional amplitude-limited coupled hyperchaotic mapping system model of the present invention; where (a) is variable g; (b) is variable a; (c) is variable b; (d) is variable c; and (e) is variable K.

[0056] Figure 3 This is a bifurcation diagram of different parameters of the three-dimensional amplitude-limited coupled hyperchaotic mapping system model of the present invention in the interval of 0 to 10; where (a) is variable g; (b) is variable a; (c) is variable b; (d) is variable c; and (e) is variable K.

[0057] Figure 4 This is a three-dimensional phase diagram and projection diagrams in three directions of the three-dimensional amplitude-limited coupled hyperchaotic mapping system model of the present invention; wherein, (a) is a three-dimensional phase diagram; (b) is a projection diagram in the xy direction; (c) is a projection diagram in the xz direction; and (d) is a projection diagram in the yz direction.

[0058] Figure 5 This is a comparison diagram of sample entropy between the three-dimensional amplitude-limited coupled hyperchaotic mapping system model of the present invention and other chaotic systems; wherein, (a) is the system used in the present invention; (b) is a two-dimensional cubic mapping; (c) is MILE; (d) is 2D-LTMM; (e) is 2D-LSMCL; (f) is 2D-CSCM;

[0059] Figure 6 This is a schematic diagram of the encryption and decryption results of the face recognition of the present invention; wherein, (a) is a plaintext image; (b) is a ciphertext image; and (c) is the decryption result. Detailed Implementation

[0060] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0061] like Figure 1 As shown, a chaotic image selective encryption method based on facial recognition and key optimization includes the following steps:

[0062] Step S1: Based on the existing chaotic mapping system model, add an exponential term and a modulo function to optimize the performance of the system model and establish a three-dimensional limited-amplitude coupled hyperchaotic mapping system model.

[0063] Step S2: Based on the three-dimensional amplitude-limited coupled hyperchaotic mapping system model, a color image face encryption scheme is designed. The specific process is as follows:

[0064] Step 21: Read the original image and extract a facial image of size M×N×3 based on the previous facial recognition results, and rearrange it to size M×3×N;

[0065] Step 22: Match the identified face image with the images in the database. If the match is successful, continue with the encryption operation.

[0066] Step 23: Use the particle swarm optimization algorithm to search for the optimal system key and control the three-dimensional limited-amplitude coupled hyperchaotic mapping system model to generate the required pseudo-random sequence;

[0067] Step 24: Perform cyclic shifting pixel rearrangement and pixel value obfuscation using half-tensor product operation on the rearranged face image, and then rearrange it to an M×N×3 size.

[0068] Step 25: Embed the encrypted face image back into the original image to obtain the final encrypted result.

[0069] Example 1

[0070] This invention discloses a chaotic image selective encryption method based on facial recognition and key optimization, which mainly consists of two parts: first, establishing a three-dimensional amplitude-limited coupled hyperchaotic mapping model, and then designing a novel face image recognition encryption scheme based on the system model.

[0071] Step S1: Establish a three-dimensional limited amplitude coupled hyperchaotic mapping system model.

[0072] Step S11: Based on the existing chaotic mapping system model, add an exponential term and a modulo function to optimize the performance of the system model; the addition of the exponential term and the modulo function plays a role in limiting the amplitude.

[0073] Based on this, a three-dimensional limited-amplitude coupled hyperchaotic mapping system model is established, as shown below:

[0074]

[0075] Where, x n y n z n e is the system's state variable; g These are the control parameters for system complexity; a, b, c, and K are conventional parameters that have little impact on system complexity. As an intermediate variable;

[0076] Step S12: Based on the established three-dimensional limited-amplitude coupled hyperchaotic mapping system model, perform dynamic behavior analysis, specifically including Lyapunov exponential diagram, bifurcation diagram, phase diagram, sample entropy, and NIST test.

[0077] The Lyapunov exponent (LE) is an important indicator for evaluating the chaotic behavior of a system. An LE value greater than 0 can be used as a reference for judging whether a dynamical system is chaotic, indicating that the system's trajectory will diverge over time. A system that meets this condition can be called a chaotic system, and the sequence generated by its iteration will also exhibit pseudo-random characteristics.

[0078] Bifurcation diagrams are a visual representation used to study dynamical systems, including chaotic systems, to illustrate how the system's behavior changes with variations in its parameters. These diagrams show how a system transitions from simple, predictable behavior to complex, chaotic dynamics through a series of bifurcations. Bifurcation occurs when a small change in parameter values ​​leads to a sudden, qualitative change in the system's behavior.

[0079] Phase diagrams for chaotic systems are key tools for describing the dynamic behavior of such systems. They graphically display the path of system state changes over time, providing an intuitive perspective for understanding the long-term behavior of the system. If the system exhibits a closed-loop trajectory, this usually indicates the presence of periodic behavior.

[0080] Sample entropy (SE), an improved version of approximate entropy (ApEn), is a statistical measure used to assess the complexity of time series data. It primarily evaluates the irregularity and complexity of time series data. Figure 5 As shown, the comparison of SE values ​​between 3D-CIMBA and other systems when varying the same range of parameters is displayed.

[0081] Table 1. NIST Test Results of the Three-Dimensional Limiting Coupled Hyperchaotic Mapping System Model

[0082]

[0083]

[0084] The NIST test can confirm whether the pseudo-random number sequence truly possesses a high level of randomness, which is crucial for ensuring the security of encryption algorithms. Using default parameters, 100 sets of data generated by the 3D amplitude-limited coupled hyperchaotic mapping were randomly selected, each set being 1 million bits long. We performed the NIST test on these datasets, and the results are shown in Table 1. The dynamic behavior analysis results of the 3D amplitude-limited coupled hyperchaotic mapping system model are as follows: Figure 2 , Figure 3 and Figure 4 As shown in the figure. The results show that, compared with other chaotic system models, the three-dimensional amplitude-limited coupled hyperchaotic mapping system model proposed in this invention has better chaotic performance, and the generated sequences exhibit good pseudo-random characteristics, which can be applied to the next step of chaotic image encryption.

[0085] Step S2: Based on the three-dimensional amplitude-limited coupled hyperchaotic mapping system model, design a color image face encryption scheme.

[0086] Step 21: Read the original image and extract a facial image of size M×N×3 based on the previous facial recognition results, and rearrange it to size M×3×N.

[0087] Step 22: Match the identified face image with the images in the database. If the match is successful, continue with the encryption operation.

[0088] Face recognition, comparison, and region extraction were performed using the FaceNet++ model. First, a face database was built using selected images. All input images to be encrypted were compared with images in the database. In the FaceNet++ model, the Euclidean distance between two images was compared to a preset threshold to determine if they belong to the same person. When the calculated maximum Euclidean distance between the input image and all face images in the database exceeded the preset threshold, a match was considered successful, and the subsequent encryption operation was performed.

[0089] Step 23: Use the particle swarm optimization algorithm to search for the optimal system key and control the three-dimensional limited-amplitude coupled hyperchaotic mapping system model to generate the required pseudo-random sequence.

[0090] Step 231: Use the particle swarm optimization algorithm to search for the optimal system key.

[0091] The basic idea of ​​the Particle Swarm Optimization (PSO) algorithm is to randomly initialize a group of particles (candidate solutions) in the search space, where each particle represents a point in the search space. Each particle has two attributes: position and velocity. Position represents a potential solution, and velocity determines the particle's direction and distance of motion. Particles update their positions by tracking two "best" solutions: one is the best position experienced by the particle (individual best solution), and the other is the best position experienced by the entire population (global best solution).

[0092] During the execution of the Particle Swarm Optimization (PSO) algorithm, the velocity and position of all particles in each generation are updated according to the following formula:

[0093]

[0094] in, Let be the particle position at time t; The position of the particle at time t+1; Let be the particle velocity vector at time t; Let p be the particle velocity vector at time t; pbest is the particle's historical best position; best iis the historical best position of the population; w is the inertia weight; c1 and c2 are the parameters that control the particle velocity; r1 and r2 are random numbers before [0,1].

[0095] In this invention, the particle dimension is set to 8 dimensions, corresponding to the 8 keys required. A 64×64×3 pixel block is randomly extracted from the image and encrypted using the keys. The peak signal-to-noise ratio and Shannon entropy of the encrypted image are calculated using formulas (4) and (5), as shown below:

[0096]

[0097] Where MSE represents the mean square error of the image; x i Input image; is the input mean; n is the number of pixels; PSNR is the peak signal-to-noise ratio.

[0098] Next, the fitness function of the particle swarm optimization algorithm is defined as follows:

[0099]

[0100] Where F(p) is the fitness; ω1 and ω2 are control coefficients, which are 0.3 and 0.7 respectively.

[0101] The quality of encryption is judged by the fitness value. After multiple iterations, an optimal solution is eventually returned, which is the required system key. Particle swarm optimization ensures that the generated sequence has good chaotic properties, thereby guaranteeing the encryption quality of subsequent encrypted images. The specific process of generating the key using the particle swarm optimization algorithm is shown in Algorithm 1.

[0102]

[0103]

[0104] Step 232: Control the three-dimensional amplitude-limited coupled hyperchaotic mapping system model to generate the required pseudo-random sequence.

[0105] Based on the system key mentioned above, generate three corresponding chaotic sequences x, y, and z. Discard the first 500 states of the sequences and use formula (7) to generate the three subsequent sequences line, row, and val:

[0106]

[0107] Among them, line i row i val i Given three distinct pseudo-random sequences; floor represents the floor function; xi y i z i is the original sequence obtained by iterating through the chaotic system; n and m are the length and width of the image to be encrypted.

[0108] Step 24: Perform cyclic shifting of pixels on the rearranged face image and obfuscate pixel values ​​using half-tensor product operation, and then rearrange it to an M×N×3 size.

[0109] Step 241: Perform pixel scrambling on the rearranged face image by alternating row and column shifts.

[0110] In the operation described in Algorithm 2, the face image is cyclically shifted using alternating rows and columns using sequences of line and row. In the subsequent encryption process, the number of iterations is set to 6000.

[0111]

[0112]

[0113] Step 242: Pixel value obfuscation based on half-tensor product operation.

[0114] If there are two matrices A∈M m×n ,B∈M p×q If n and p have a least common multiple of t, then T is defined as the semi-tensor product of the two matrices, as shown below:

[0115]

[0116] Where M represents the set of all matrices of size m×n; ∝ represents the semi-tensor product operation; represents the tensor product of matrices; I represents the identity matrix.

[0117] Step 25: Embed the encrypted face image back into the original image to obtain the final encryption result; The diagram illustrating the face recognition encryption and decryption result is shown below. Figure 6 As shown.

[0118] Example 2

[0119] This embodiment tests the facial database comparison function.

[0120] The facial test set was obtained from the Internet Movie Database (IMDB). Due to the large size of the files, 3000 images from IMDB.0, totaling 1.3GB, were used to build the facial database.

[0121] The following five images (rm342264064_1964-9-2_2009.jpg, rm304650752_1955-5-16_2001.jpg, rm3388605-44_1961-7-30_1999.jpg, rm3794573056_1985-7-18_2007.jpg, rm848858368_1978-9-23_2010.jpg) are used as input. For convenience, these will be abbreviated as "Images 1 to 5". Four images successfully match the images with the smallest Euclidean distance in the database, while one fails, as shown in Table 2. Encryption is then performed on these four successfully matched face images.

[0122] Table 2. Matching results of different face recognition methods

[0123]

[0124] Example 3

[0125] This embodiment provides a specific example of a semi-tensor product operation: Assume there are two matrices, P∈M 3×4 ,B∈M 2×3 ,

[0126]

[0127] On the contrary, there are

[0128]

[0129] Then, Q can be easily recovered from the equation. It is worth noting that, since P is not a square matrix, operations on P... -1 This corresponds to finding its generalized inverse matrix. In the encryption steps of this invention, a half-tensor product with multiple dimensions is typically used to simplify the calculation, i.e., the special case where n is an integer multiple of p.

[0130] In encryption, extract the first n bits of val. 2 Each element is used to construct a square matrix R. nxn Then, calculate R. nxn The final encrypted result, cipher, is obtained by taking the half-tensor product of the disordered image and then dividing all elements of the matrix by 256 and taking the remainder.

[0131] Because R needs to be calculated during the decryption process. nxn The inverse matrix of R is obtained, therefore it must be ensured that the matrix is ​​invertible. If the first generated R... nxnIf it's not reversible, then we should start taking values ​​from the 101st element of `val` to regenerate it. If it's still not reversible, we continue taking values ​​from every subsequent 100 elements until a reversible full-rank matrix is ​​produced. At this point, we obtain the final encryption result. Since the encryption scheme proposed in this invention is a symmetric encryption, the decryption process essentially involves the reverse operation of the encryption process.

[0132] Therefore, this invention employs the aforementioned chaotic image selective encryption method based on facial recognition and key optimization, establishing a three-dimensional amplitude-limited coupled hyperchaotic mapping system model. This system model has a controllable Lyapunov exponent, a wide chaotic range, high complexity, and selectively encrypts facial regions. A face recognition encryption scheme based on the DeepFace model is adopted to effectively improve encryption speed. The PSO optimization algorithm is used to control the initial values ​​and control parameters generated in the system, generating a key associated with the image to be encrypted, enabling the encryption scheme to effectively resist known plaintext attacks. At the same time, STP, as a novel scheme for solving pixel obfuscation in images, is significantly different from traditional XOR-based obfuscation methods, making it effective against traditional attack methods and offering advantages in computational efficiency and encryption quality.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for selective encryption of chaotic images based on facial recognition and key optimization, characterized in that, Includes the following steps: Step S1: Based on the existing chaotic mapping system model, an exponential term and a modulo function are added to optimize the system model's performance. The addition of the exponential term and the modulo function serves to limit the amplitude. Based on this, a three-dimensional amplitude-limited coupled hyperchaotic mapping system model is established, as shown below: （1）； in, , , These are the state variables of the system; These are control parameters for system complexity; , , , These are standard parameters; As an intermediate variable; Based on the established three-dimensional limited-amplitude coupled hyperchaotic mapping system model, dynamic behavior analysis is carried out, specifically including Lyapunov exponential diagram, bifurcation diagram, phase diagram, sample entropy, and NIST test; Step S2: Based on the three-dimensional amplitude-limited coupled hyperchaotic mapping system model, a color image face encryption scheme is designed. The specific process is as follows: Step 21: Read the original image and extract a facial image of size M×N×3 based on the previous facial recognition results, and rearrange it to size M×3×N; Step 22: Match the identified face image with the images in the database. If the match is successful, continue with the encryption operation. Step 23: Use the particle swarm optimization algorithm to search for the optimal system key and control the three-dimensional limited-amplitude coupled hyperchaotic mapping system model to generate the required pseudo-random sequence; Step 24: Perform cyclic shifting pixel rearrangement and pixel value obfuscation using half-tensor product operation on the rearranged face image, and then rearrange it to an M×N×3 size. Step 25: Embed the encrypted face image back into the original image to obtain the final encrypted result.

2. The chaotic image selective encryption method based on facial recognition and key optimization according to claim 1, characterized in that: In step S22, the FaceNet++ model is used to perform face recognition, comparison, and region extraction. First, a face database is built using selected images, and all input images to be encrypted are compared with the images in the database. Then, in the FaceNet++ model, the Euclidean distance between two images is compared with a preset threshold to determine whether they belong to the same person. A match is considered successful when the calculated maximum Euclidean distance between the input image and all face images in the database is greater than a preset threshold.

3. The chaotic image selective encryption method based on facial recognition and key optimization according to claim 1, characterized in that, In step S23, the optimal system key is searched using the particle swarm optimization algorithm, and the three-dimensional limited-amplitude coupled hyperchaotic mapping system model is controlled to generate the required pseudo-random sequence. The specific process is as follows: Step 231: Use the particle swarm optimization algorithm to search for the optimal system key; During the execution of the particle swarm optimization algorithm, the velocity and position of all particles in each generation are updated according to the following formula: （2）； （3）； in, Let be the particle position at time t; The position of the particle at time t+1; Let be the particle velocity vector at time t; Let be the particle velocity vector at time t+1; This is the best historical position for the particle; This is the best historical position for the population; Inertial weight; , To maintain the particle velocity; , A random number before [0,1]; The particle dimension is set to 8, corresponding to the 8 keys required; a 64×64×3 pixel block is randomly extracted from the image and encrypted using the keys. The peak signal-to-noise ratio and Shannon entropy of the encrypted image are calculated using formulas (4) and (5), as shown below: （4）； （5）； in, This represents the mean square error of the image; Input image; Input the mean; Number of pixels; Peak signal-to-noise ratio; Next, the fitness function of the particle swarm optimization algorithm is defined as follows: （6）； in, For fitness; , The control factors are set to 0.3 and 0.7 respectively. The quality of encryption is judged by the size of this fitness. After multiple rounds of iteration, an optimal solution is returned, which is the required system key. Step 232: Control the three-dimensional amplitude-limited coupled hyperchaotic mapping system model to generate the required pseudo-random sequence; Based on the system key mentioned above, generate three corresponding chaotic sequences x, y, and z; discard the first 500 states of the sequence, and use formula (7) to generate the three subsequent sequences. , and : （7）； in, , , These are three different pseudo-random sequences; The function that represents rounding down; , , The original sequence obtained by iterating through the chaotic system; , The length and width of the image to be encrypted.

4. The chaotic image selective encryption method based on facial recognition and key optimization according to claim 1, characterized in that, In step S24, the rearranged face image undergoes cyclic shifting pixel rearrangement and pixel value obfuscation using half-tensor product operation, and then it is rearranged to a size of M×N×3. The specific process is as follows: Step 241: Scramble the pixels of the rearranged face image by alternating row and column shifts; Step 242: Pixel value obfuscation based on half-tensor product operation; There are two matrices If n and p have a least common multiple of t, then T is defined as the semi-tensor product of the two matrices, as shown below: （8）； Where M represents all sizes of The set of matrices; This represents the operation of half-tensor product; Represents the tensor product of matrices; Represents the identity matrix.

Images