A selective encryption method and system based on a fibonacci dynamic diffusion algorithm and semantic perception fusion

Abstract

This invention relates to a selective encryption method and system based on the Fibonacci dynamic diffusion algorithm and semantic awareness. The method includes: acquiring an original color image; using a semantic awareness-driven hierarchical encryption mechanism to identify the importance of each region in the original color image, obtaining important region images and secondary important region images; performing bit-level cross-scrambling on the important region images to generate a three-dimensional tensor, and dynamically diffusing the important region images based on a dynamic diffusion enhancement mechanism to obtain dynamically diffused important region images; the dynamic diffusion enhancement mechanism is obtained by introducing position-related dynamic parameters into the original dynamic diffusion enhancement mechanism; and performing global chaotic block scrambling and chain-like diffusion encryption on the dynamically diffused important region images and the secondary important region images to obtain the final encrypted image.

Description

Technical Field

[0001] This invention relates to the field of image security technology, and in particular to a selective encryption method and system based on the Fibonacci dynamic diffusion algorithm and semantic awareness. Background Technology

[0002] With the rapid advancement of communication technology, the internet has deeply integrated into all aspects of social production and life. Images, due to their vivid and intuitive characteristics, have become one of the most important carriers of information transmission. However, the high frequency of image applications harbors multiple security risks; subtle differences in medical images can directly affect diagnostic accuracy. In various scenarios, ensuring image security has become a critical and urgent technical task. To address the complex challenges in the field of image security, existing technical solutions have established a multi-layered policy system combining laws, regulations, and standards. However, due to the characteristics of image data, such as large data volume, high redundancy, and strong pixel correlation, traditional encryption algorithms (such as AES, DES, and RSA) often struggle to simultaneously meet the requirements of security and real-time performance.

[0003] Furthermore, existing technologies often employ chaotic systems with fixed parameters, where the keystream is independent of the plaintext, making them vulnerable to chosen-plaintext attacks. Global encryption is secure but inefficient, while traditional selective encryption is efficient but its security is questionable. Many encryption schemes use static or fixed parameters. The diffusion process follows a fixed transformation rule, which reduces the algorithm's nonlinear complexity and resistance to analysis.

[0004] Furthermore, research hotspots both domestically and internationally focus on designing more complex chaotic systems and introducing novel computational models to enhance security. Due to its high nonlinearity and complexity, high computational efficiency, and reversibility, it provides excellent cryptographic properties such as pseudo-randomness, avalanche effect, and sensitivity to initial conditions for encryption algorithms. It has been widely used by many scholars for image encryption. However, most of the existing mainstream encryption algorithms adopt "full-image equal-strength encryption", which faces bottlenecks of low computational efficiency and high resource consumption when processing high-resolution images, making it difficult to meet the needs of real-time communication.

[0005] Therefore, a novel approach to selective encryption based on the Fibonacci dynamic diffusion algorithm and semantic awareness is urgently needed. By distinguishing the importance of image content, "precise encryption" is implemented, significantly improving encryption efficiency without sacrificing core information security. Simultaneously, image features are extracted using SHA3-512 hashing to generate an initial key for a 2D-HNN chaotic system strongly correlated with the plaintext, achieving "one-time pad" encryption and maximizing the advantages of semantically aware layered image encryption. This approach has significant theoretical and practical value in promoting the development of efficient multimedia security technologies. Summary of the Invention

[0006] To address the problems existing in the prior art, the present invention aims to provide a selective encryption method and system based on the Fibonacci dynamic diffusion algorithm and semantic awareness. This method achieves multi-strategy precise image encryption by constructing a hierarchical encryption framework that integrates semantic awareness. It extracts high-dimensional image features based on SHA3-512 to generate an initial key, which drives a 2D-HNN chaotic system to produce highly random and high-dimensional chaotic sequences. This significantly enhances the unpredictability and sensitivity of the key, exhibiting significant advantages in resisting known / chosen plaintext attacks while maintaining good diffusion and robustness, and simultaneously improving security. Furthermore, it introduces a position-dependent dynamic exponential parameter (i.e., matrix exponentiation). (Dynamically changing with chaotic sequences) causes different pixel pairs in the image to undergo linear transformations of different topological structures, thus completely breaking the periodicity limitation of traditional matrix transformations. While maintaining computational efficiency, it greatly improves the nonlinear complexity and resistance to differential attacks of the algorithm.

[0007] To achieve the above objectives, the present invention provides the following solution: A selective encryption method based on the Fibonacci dynamic diffusion algorithm and semantic awareness is proposed, comprising: The original color image is acquired, and the importance of each region in the original color image is identified by a semantically perceptual-driven hierarchical encryption mechanism to obtain images of important regions and images of less important regions. The image of the important region is subjected to bit-level cross-scrambling to generate a three-dimensional tensor, and based on... The dynamic diffusion enhancement mechanism dynamically diffuses the important region image to obtain a dynamically diffused important region image; the dynamic diffusion enhancement mechanism is obtained by introducing position-related dynamic parameters into the original dynamic diffusion enhancement mechanism. Global chaotic block scrambling and chain-diffusion encryption are performed on the important region image and the secondary important region image after dynamic diffusion to obtain the final encrypted image.

[0008] Optionally, the method further includes: The original color image is subjected to dimensional constraint preprocessing and reshaped into a one-dimensional column vector; The one-dimensional column vector is input into the target hash function to generate a fixed-bit hash value, which is then converted into a target-bit unsigned integer numerical vector and mapped to a hash string composed of multiple hexadecimal characters. The hash string is converted into a decimal feature vector, which is used to calculate the initial state value, gain parameter, and weight matrix parameter of the 2D-HNN hyperchaotic system in order to construct a two-dimensional discrete chaotic model. The initial state is input into the two-dimensional discrete chaotic model, and after multiple pre-iterations to eliminate transient effects, iterative extraction continues. The component sequences are used to obtain the core one-dimensional sequence, and the PCHIP operator is used to perform nonlinear fitting on the core points to generate a full-length global chaotic sequence of the target length; the target length is the required total length estimated after multiple pre-iterations to eliminate transient effects. The full-length global chaotic sequence is divided into bit scrambling sequences. Diffusion sequence, global block scrambling sequence, and global diffusion parameters.

[0009] Optionally, obtaining the important region image and the secondary important region image includes: The original color image is converted to a grayscale image, using... The operator performs a convolution operation on the grayscale image and calculates the horizontal and vertical gradients respectively. The combined gradient magnitude of each pixel is calculated using the horizontal and vertical gradients to construct a semantic gradient map. The semantic gradient map is flattened into a one-dimensional column vector and sorted in descending order of numerical value to obtain the corresponding original position index vector. The original color image is reshaped into a two-dimensional matrix, the row vectors of the two-dimensional matrix are rearranged according to the order of the original position index vectors, and then reshaped into the target three-dimensional tensor to obtain a semantically ordered image. The semantically ordered image is segmented according to the segmentation threshold to obtain the important region image and the secondary important region image.

[0010] Optionally, generating the three-dimensional tensor includes: The important region image is reduced in dimension and reshaped into a one-dimensional pixel column vector. Bit extraction is then used to decompose each pixel in the one-dimensional pixel column vector into multiple binary bits to construct a global bit vector. The bit scrambling sequence is sorted in ascending order to generate a global position scrambling index vector, which is used to perform position mapping on the global bit vector to generate a scrambled bit vector. The scrambled bit vector is recombined into decimal pixel values ​​and summed according to binary weight rules to obtain a recombined pixel vector. The recombined pixel vector is then reverse-engineered into the three-dimensional tensor.

[0011] Optionally, obtaining the image of the important region after dynamic diffusion includes: The three-dimensional tensor is converted into double-precision floating-point data to perform a dimension reshaping operation, which flattens the three-dimensional tensor and re-divides it into a pixel pair matrix. use The diffusion sequence maps chaotic sequence values ​​to a dynamic exponential sequence within an integer range, and confuses adjacent pixel vectors in the pixel pair matrix by performing a matrix power operation on the number of transformations corresponding to the dynamic exponential sequence, thereby generating a transformed pixel pair matrix. The transformed pixel pair matrix is ​​inversely reshaped to its original size and converted to an unsigned integer of the target bit to obtain the image of the important region after dynamic diffusion.

[0012] Optionally, performing the global chaotic block scrambling includes: The important region image and the secondary important region image after dynamic diffusion are tensor-stitched in the horizontal direction to reconstruct a full-size intermediate image, and then dimensionality-reduced and reshaped into a two-dimensional pixel matrix. The two-dimensional pixel matrix is ​​divided into multiple independent image blocks with the goal of treating multiple adjacent pixels as a single encryption unit, in order to construct an original pixel index sequence, and then the original pixel index sequence is reshaped into a block index matrix. The global block scrambling sequence is sorted in ascending order to obtain the block position permutation index vector. The block position permutation index vector is then used to perform a global permutation on the column vectors of the block index matrix to obtain the scrambled index matrix. The scrambled index matrix is ​​flattened into a one-dimensional index vector, which is used to extract pixel data from the two-dimensional pixel matrix, construct a scrambled two-dimensional matrix, and then reshape it into a three-dimensional image tensor of the target size.

[0013] Optionally, obtaining the final encrypted image includes: The three-dimensional image tensor is reduced and reshaped into a one-dimensional pixel input sequence, and a nonlinear quantization operation is performed using a global diffusion chaotic sequence to adapt to the target pixel position, thereby generating a diffusion key stream. The arithmetic-logic-displacement ternary hybrid diffusion model is used to break the linear mapping law through key-controlled dynamic displacement to forward diffuse the diffusion key stream and generate a one-dimensional ciphertext vector. The one-dimensional ciphertext vector is inversely reshaped into a three-dimensional image tensor and then converted into an unsigned integer of the target bit to obtain the final encrypted image.

[0014] Optionally, generating the one-dimensional ciphertext vector includes: ; in, It is a one-dimensional ciphertext vector. This means using an 8-bit binary number Move left in a loop Bit, , For diffusion of key streams, The first pixel in the current input one-dimensional pixel sequence or the scrambled pixel sequence. pixel value, The ciphertext pixel values ​​obtained from the previous loop calculation. Modulo operation is used to truncate and restrict a value to a certain range. Within a single byte range.

[0015] To achieve the above objectives, the present invention also provides a selective encryption system based on the Fibonacci dynamic diffusion algorithm and semantic awareness, comprising: The data acquisition module is used to acquire raw color images; The semantic awareness module is used to identify the importance of each region in the original color image using a semantic awareness-driven hierarchical encryption mechanism, and to obtain images of important regions and images of less important regions. The dynamic diffusion module is used to perform bit-level cross-scrambling on the image of the important region, generate a three-dimensional tensor, and based on... The dynamic diffusion enhancement mechanism dynamically diffuses the important region image to obtain a dynamically diffused important region image; the dynamic diffusion enhancement mechanism is obtained by introducing position-related dynamic parameters into the original dynamic diffusion enhancement mechanism. The diffusion encryption module is used to perform global chaotic block scrambling and chain diffusion encryption on the important region image and the secondary important region image after dynamic diffusion to obtain the final encrypted image.

[0016] The beneficial effects of this invention are as follows: This invention uses a semantic importance recognition mechanism to intelligently segment and differentially encrypt image content, providing high-strength protection for important areas and lightweight processing for less important areas. This effectively improves overall encryption efficiency while ensuring security, and has good practicality and reasonable resource allocation.

[0017] This invention introduces The dynamic diffusion mechanism allows the diffusion rules to change dynamically with the key, significantly enhancing the system's resistance to traditional cryptographic analysis methods such as statistical analysis and differential attacks, and providing higher security robustness and adaptability.

[0018] This invention supports semantic segmentation and dynamic parameter adjustment in its module design. It can implement differentiated encryption strategies for different image content and improve the overall security strength through dynamic diffusion mechanism. It has good scalability and applicability and is suitable for image encryption scenarios with high requirements for both security and efficiency.

[0019] This invention optimizes perceptual encryption from multiple perspectives by fusing the SHA3-512 hash algorithm with a 2D-HNN chaotic system. It deeply extracts high-dimensional features from images using the SHA3-512 hash algorithm and generates an initial key strongly correlated with the plaintext. This key drives a high-performance 2D-HNN chaotic system, constructing a chaotic sequence with extremely high randomness and complexity. This mechanism fundamentally solves the inherent vulnerabilities of traditional fixed-parameter chaotic systems, such as limited key space and susceptibility to known / chosen plaintext attacks. While ensuring excellent diffusion effects and algorithm robustness, it achieves a highly efficient improvement in plaintext sensitivity and encryption unpredictability.

[0020] In summary, this invention, through its two core mechanisms of semantically aware layered encryption and dynamic parameter diffusion, demonstrates significant advantages in terms of structural rationality, security strength, and execution efficiency, and possesses strong practical value and technological foresight. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness, according to an embodiment of the present invention. Figure 2 Embodiments of the present invention Dynamic diffusion diagram; Figure 3 This is a flowchart illustrating the decryption process of an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the execution principle of the semantic awareness and adaptive block segmentation algorithm in an embodiment of the present invention. Figure 5 This is an evolution diagram of the distribution features of candy images before and after semantic-aware segmentation in an embodiment of the present invention; Figure 6 This is an evolution diagram of the distribution features of complex natural landscape images before and after semantic perception segmentation in an embodiment of the present invention; Figure 7 This is a comparison chart of the statistical characteristics of the entire encryption and decryption process of four benchmark dataset images in this embodiment of the invention; Figure 8 This is a sensitivity evaluation chart for differential attack resistance of different reference color images in three channels according to an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Terminology Explanation: (1) SHA3-512: The third generation of secure hash algorithm standard, which outputs a 512-bit hash value for data integrity verification.

[0025] (2) 2D-HNN (two-dimensional hyperchaotic neural network): a dynamic system with high-dimensional chaotic characteristics, used to generate pseudo-random sequences required for encryption.

[0026] (3) The nth power of Q, a 2×2 matrix constructed from the Fibonacci sequence, is used in this invention for the diffusion transformation of pixel values.

[0027] (4) Operator: The operator combines Gaussian smoothing and differential differentiation, and contains two... The convolutional kernels are used to detect edge responses in the horizontal and vertical directions, respectively. The horizontal convolutional kernel is defined. Convolution kernel in the vertical direction as follows: ; Using the above convolution kernels in grayscale images Perform convolution operations on the pixel to calculate the gradient component in the horizontal direction for each pixel. and the gradient component in the vertical direction : ; ; in, This represents a two-dimensional discrete convolution operation.

[0028] (5) PCHIP interpolation algorithm: A small number of core skeleton points are calculated through the 2D-HNN dynamic equation. The piecewise cubic Hermitian interpolation polynomial (PCHIP) is used to construct a smooth and monotonic chaotic trajectory between the skeleton points, thereby quickly expanding and generating the full-length key stream.

[0029] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] like Figure 1 As shown, this embodiment discloses a selective encryption method based on the Fibonacci dynamic diffusion algorithm and semantic awareness, comprising: acquiring an original color image; using a semantic awareness-driven hierarchical encryption mechanism to identify the importance of each region in the original color image, obtaining important region images and secondary important region images; performing bit-level cross-scrambling on the important region images to generate a three-dimensional tensor, and based on... The dynamic diffusion enhancement mechanism dynamically diffuses the important region image to obtain the dynamically diffused important region image; the dynamic diffusion enhancement mechanism is obtained by introducing position-related dynamic parameters into the original dynamic diffusion enhancement mechanism; global chaotic block scrambling and chain diffusion encryption are performed on the dynamically diffused important region image and the secondary important region image to obtain the final encrypted image.

[0031] Specifically, this embodiment discloses a selective encryption method based on the Fibonacci dynamic diffusion algorithm and semantic awareness, comprising: Step 1. Plaintext association key generation and chaotic sequence preparation based on SHA3-512: (1) Image preprocessing and hash calculation: Assume the input original color image is P, with size... (in First, perform dimensional constraint preprocessing on P: if H or W is odd, adjust it to an even number through a cropping operation. Let the processed image data be denoted as . To achieve a dynamic encryption mechanism of "one-time key", the following will be implemented: Reshape into a one-dimensional column vector , and enter to The hash function generates a 512-bit (64-byte) hash value. This hash value is first converted into an 8-bit unsigned integer vector, and then mapped to a hash string consisting of 128 hexadecimal characters. .

[0032] (2) Quantization mapping of initial parameters of chaotic systems: Will Convert to a decimal feature vector : ; in For the first The vector contains decimal values ​​corresponding to each hexadecimal character. This will serve as the sole seed for generating subsequent dynamic parameters.

[0033] To overcome the problems of insufficient uniformity of initial value distribution and weak nonlinearity caused by traditional linear modulo methods, this invention proposes a nonlinear mapping algorithm based on XOR avalanche and logarithmic modulation. Based on vector... The initial state values ​​of the 2D-HNN hyperchaotic system are calculated using the following formula. Gain parameter β and weight matrix parameters The specific definition formula is as follows: ; ; for Introducing the natural logarithm function Increasing the transcendence of numerical values, for Introducing the tangent function Increase the system's divergence and combine it with the bitwise XOR operation of adjacent elements ( Introducing logical nonlinearity, finally 1. Ensure the values ​​strictly fall within the range of Within the chaotic attraction domain.

[0034] ; Normalize the complex summation results to The range is defined, and it exhibits extremely high nonlinearity. A further linear transformation is performed to constrain the gain parameter β within a certain range. Within the interval, ensure that the HNN neural network operates in a highly chaotic, high-gain region.

[0035] ; ; ; ; in, , Hyperbolic tangent function. Weight parameters. The strength of the nonlinear interaction between neurons is determined, and its value range is precisely mapped to optimize the distribution characteristics of the chaotic attractor.

[0036] Step 2.2 D-HNN Chaotic Sequence Generation and Interpolation Optimization: This invention employs a two-dimensional discrete chaotic equation with high gain characteristics: ; in, This represents the number of iterations. Initial state. , To begin the iteration (when) Substitute the initial value into the equation (at time), and the gain parameter Weight matrix parameters , .

[0037] This two-dimensional discrete chaotic equation generates chaotic values ​​with extremely high randomness, unpredictability, and extreme sensitivity to initial conditions through nonlinear iteration. These values ​​serve as the source of the secure key stream for all subsequent encryption operations (bit scrambling, dynamic Fibonacci obfuscation, block scrambling, and chain diffusion) and are also used to compute the core one-dimensional sequence.

[0038] To balance encryption security and system execution efficiency, this invention introduces a piecewise cubic Hermitian (PCHIP) interpolation algorithm. First, 500 pre-iterations are performed to eliminate transient effects. Then, the required total length is estimated. : ; The core one-dimensional sequence is calculated using an interpolation factor of Rate=16. Specifically, the initial state and Substituting into the above two-dimensional discrete chaotic equation, after 500 pre-iterations to eliminate transient effects, the system is iterated further to obtain the system's... Component sequences, thus obtaining the core one-dimensional sequence. (i.e., the core point). Subsequently, the PCHIP operator is used to perform nonlinear fitting on the core point to generate a full-length global chaotic sequence. The formula is shown below: ; Finally Cut into , , , Four sequences, used as chaotic sequences for subsequent encryption, have lengths of respectively , , , The sequence lengths and segmentation formulas for each module are defined as follows: ; ; ; ; in It is the length of the bit scrambling sequence. Diffusion sequence length, This refers to the length of the global block scrambling sequence. This refers to the length of the global diffusion parameter.

[0039] ; ; ; ; in It refers to a bit scrambling sequence. diffusion sequence, This refers to the global block scrambling sequence. It refers to a globally diffused chaotic sequence.

[0040] Step 3. Semantic importance calculation and adaptive block segmentation, such as Figure 4 As shown: (1) Construction of semantic gradient graph: First, replace P with a grayscale image. To eliminate interference from color components. Then, using... Operator pairs Perform convolution operations and calculate the horizontal gradient separately. and vertical gradient Based on this, the comprehensive gradient magnitude of each pixel is calculated to construct a semantic gradient map. The calculation formula is as follows: ; in, Represents pixel coordinates, The value of directly reflects the texture richness and edge strength (i.e. semantic importance) of the pixel at that location.

[0041] (2) Pixel-level semantic sorting: To distinguish between important and less important information, the system... Perform vectorization and sorting operations. Transform the 2D gradient map. Flatten the vector into a one-dimensional column vector, sort it in descending order of its values, and obtain the corresponding original position index vector. The mathematical expression is: ; in, It is an index sequence containing the original position information of pixels. The indexes at the beginning of the sequence correspond to high-frequency detail regions (such as edges and contours) in the image, and the indexes at the end correspond to low-frequency smooth regions (such as the background).

[0042] (3) Image reconstruction and adaptive segmentation: Using index vectors The pixel data of the original image P is remapped. First, the original image P is reshaped into a two-dimensional matrix. Next, according to Rearrange the order The row vectors; finally, the sorted data is reshaped into... The three-dimensional tensor is used to obtain a semantically ordered image. At this time, high gradient pixels are concentrated in The left column contains low-gradient pixels, while the right column contains low-gradient pixels. The segmentation threshold is set to half the image width. ,Will Vertically divided into two parts: Important areas ( ):extract The former The column contains approximately 50% of the pixels with high semantic information in the entire image.

[0043] Secondary important areas ( ):extract After The column contains the remaining pixels with low semantic information.

[0044] The segmentation formula is defined as follows: ; In one possible implementation, a test image of a typical object (candy / bean) is used as an example to demonstrate the changes in statistical features before and after pixel transfer. Figure 5 The original plaintext and its color histogram were compared, along with the extracted top 50% high-frequency important regions and the bottom 50% low-frequency secondary important regions. Because the remapping of spatial location features disrupts the correlation between adjacent pixels in the original image, the histograms of important regions exhibit a complex, broad distribution (clustering edge textures), while the histograms of secondary important regions show extremely high peak clustering (smoothing the background). This demonstrates that the algorithm can accurately extract semantic features at the statistical level.

[0045] In another possible implementation, a test image of a complex natural landscape (trees and coastline) is used as an example to further verify the universality of the semantic perception mechanism. Under more complex natural textures, after descending order rearrangement and truncation, the important regions still perfectly retain almost all color abrupt changes and high-frequency components, visually presenting a high-density noise state; while the less important regions are almost washed away to a single background color distribution, such as... Figure 6 As shown.

[0046] Step 4. Bit-level cross-scrambling of critical regions: (1) Image vectorization and bit-plane decomposition: Image of important regions The size is The total number of pixels is First, Dimensionality reduction and reshaping into a one-dimensional pixel column vector Then, bit extraction is used to extract each pixel. It is decomposed into 8 binary bits. Next, a global bit vector is constructed. , of which The first pixel Bits Mapped to a vector Specific location.

[0047] The decomposition formula is as follows: ; in, This indicates the floor function. For pixel index, Bit index ( Indicates the least significant bit (LSB). (Indicates the most significant bit, MSB).

[0048] (2) Global bit scrambling based on chaotic sequences: The length generated in the Step 2 preparation stage is chaotic sequence Through the analysis of Sort in ascending order and generate a global position scrambling index vector. : ; in It is a collection of arrive A vector of random permutations of all integers.

[0049] Based on this index vector, the global bit vector Perform position mapping to generate a scrambled bit vector. : ; This operation allows any bit of any pixel to be swapped to any bit position of any other pixel in the image, achieving complete bit-level cross-scrambling.

[0050] (3) Pixel-weighted reassembly and image reconstruction: The scrambled bit vector Recombined into decimal pixel values. For the first... Recombined pixels The summation is performed according to the binary weighting rules: ; Finally, it will contain all recombined pixels. Inverse reshaping into a three-dimensional tensor This serves as the input for subsequent dynamic Fibonacci diffusion.

[0051] Step 5. Important Areas Dynamic diffusion, such as Figure 2 As shown: (1) Pixel vectorization and grouping: Obtained from step 4 The size is First, The data is converted to double-precision floating-point number for numerical computation. Then, a dimension reshaping operation is performed to transform the resulting three-dimensional tensor... Flatten directly and re-divide into pixel pair matrix Each column represents a pair of adjacent pixel vectors. .

[0052] ; This operation couples spatially adjacent pixels into independent transformation units.

[0053] (2) Generation of dynamic transformation index: Using the chaotic sequence generated in Step 2 To control the diffusion intensity. A nonlinear quantization formula maps chaotic sequence values ​​to an integer range. Dynamic exponential sequence within .

[0054] For the Group of pixel pairs, the number of transformations they correspond to The calculation is as follows: ; in, The range of values ​​ensures that the transformation matrix has sufficient confusion and that the computational cost is controllable.

[0055] (3) Dynamic Matrix transformations: Basic Fibonacci transformation matrix : ; For each pixel pair vector ,application Obfuscation is applied to the exponentiation of the matrix. To prevent pixel values ​​from overflowing, the operations are performed in the modulo 256 domain.

[0056] The transformation formula is defined as follows: ; Right now: ; (4) Image reconstruction: Transformed pixel pair matrix Remodeling to original size The data is then converted to an 8-bit unsigned integer (uint8) to obtain the image of the important region after dynamic diffusion. .

[0057] This step involves introducing location-related dynamic parameters. Breaking with traditional Arnold or The fixed transformation period significantly increases the nonlinear complexity of the algorithm.

[0058] Step 6. Scramble the global chaos block: (1) Region fusion and tensor dimensionality reduction: The image of the important region after dynamic diffusion is The image of the secondary important region is First, tensor stitching is performed on the two parts horizontally to reconstruct an intermediate image of full size. : ; To facilitate block-level operations, the 3D image tensor Dimensional reduction and reshaping into a two-dimensional pixel matrix In this matrix, each row represents the color vector of a pixel (containing R, G, and B components), and each column represents a color channel.

[0059] ; (2) Construction of pixel index block matrix: Define block size as (That is, every 4 adjacent pixels form an encryption unit), then the two-dimensional pixel matrix Classified as Individual image patches, in which .

[0060] Construct the original pixel index sequence and reshape it into Block index matrix In this matrix, the first... Column vectors represent the first The original position indices of the 4 pixels contained in each image patch: .

[0061] (3) Block-level permutations based on chaotic sequences: Scramble the chaotic sequence using the global block generated in Step 2 (length is) Sort the sequence in ascending order and obtain the block position permutation index vector. : ; in yes arrive A random arrangement.

[0062] Using indexes For the block index matrix The column vectors are permuted to obtain the disordered index matrix. : ; This operation enables random position swapping of image patches across the entire image, rather than being limited to local areas.

[0063] (4) Image remapping and reconstruction: Disordered index matrix Flattened into a one-dimensional index vector Using this vector from Extract pixel data and construct a scrambled two-dimensional matrix. : ; Finally, Reverse reshaping 3D image tensor Complete the global chaotic block scrambling.

[0064] Step 7. Chain diffusion encryption based on hybrid domain dynamic cyclic shift: (1) Diffusion domain vectorization and key stream adaptation: First, after the Step 6 block scrambling process... Dimensional reduction and reshaping into a one-dimensional pixel input sequence ,in This represents the total number of pixels in the image. The globally diffused chaotic sequence generated in Step 2 is invoked synchronously. Perform non-linear quantization to fit the 8-bit pixel, generating a diffusion key stream. : ; in, Indicates the first A diffusion key, To act as a magnification factor to preserve minute details of the chaotic sequence, This is for floor function.

[0065] (2) Dynamic cyclic shift-diffusion equation of the mixed domain: for For sequential pixels, a ternary hybrid diffusion model of arithmetic-logic-displacement is constructed. This model forces the data stream to continuously traverse different algebraic groups, using key-controlled dynamic displacement to break the linear mapping law. The forward diffusion equation is defined as follows: ; Among them, the circular left shift operator This means using an 8-bit binary number Move left in a loop Bit, .

[0066] (3) Ciphertext Reconstruction: After completing the diffusion operation of the full-length sequence, the one-dimensional ciphertext vector is... The image is inversely reconstructed into a 3D image tensor and then converted to an 8-bit unsigned integer (uint8) to obtain the final encrypted image. : .

[0067] Figure 7 The system's encryption effectiveness against four different types of benchmark images is demonstrated. The charts employ a full matrix comparison, including plaintext, ciphertext, the actual decrypted and recovered images, and their respective RGB three-channel color histograms. Regardless of how rugged the histogram distribution of the original plaintext is (like mountain peaks), after the chaotic block scrambling and three-dimensional cross-diffusion of this invention, the histogram of the ciphertext image is flattened, presenting an extremely perfect "uniform distribution with equal probability." This theoretically completely cuts off the path for attackers attempting to crack the key through statistical frequency analysis, and the decryption process achieves lossless recovery with zero data loss. The decryption process, as shown... Figure 3 As shown.

[0068] Figure 8 This chart presents the core performance metrics against plaintext differential attacks. The chart uses a split-column bar chart, plotting the pixel change rate (NPCR) and normalized average change intensity (UACI) for four images across independent R, G, and B channels. Because this invention uses SHA3-512 deep hashing to extract image features and generate associated keys, even a single bit change in the plaintext will trigger an avalanche effect. The NPCR of all test data closely approximates the theoretical limit of 99.6094%, and the UACI closely approximates the theoretical limit of 33.4635%, demonstrating that the system possesses top-tier security sensitivity against known-plaintext attacks (KPA) and chosen-plaintext attacks (CPA).

[0069] This embodiment also provides a selective encryption system based on the Fibonacci dynamic diffusion algorithm and semantic awareness, including: a data acquisition module for acquiring original color images; and a semantic awareness module for using a semantic awareness-driven hierarchical encryption mechanism to identify the importance of each region in the original color image and obtain images of important regions and images of less important regions. The dynamic diffusion module is used to perform bit-level cross-scrambling on the image of the important region, generate a three-dimensional tensor, and based on... The dynamic diffusion enhancement mechanism dynamically diffuses the important region image to obtain the dynamically diffused important region image; the dynamic diffusion enhancement mechanism is obtained by introducing position-related dynamic parameters into the original dynamic diffusion enhancement mechanism; the diffusion encryption module is used to perform global chaotic block scrambling and chain diffusion encryption on the dynamically diffused important region image and the secondary important region image to obtain the final encrypted image.

[0070] This invention constructs an image semantic importance calculation module to achieve intelligent semantic parsing of uploaded images, thereby accurately identifying the importance of content in each region. Specifically, the image is first divided into multiple pixel blocks, and then categorized into "important regions" and "secondary important regions" based on semantic importance. This transforms the traditional globally unified encryption method into a targeted, layered encryption strategy. For example, more complex encryption modules are used for important image blocks, while secondary important blocks undergo lightweight processing, significantly improving computational efficiency while maintaining security strength. This mechanism achieves effective synergy between security and processing efficiency by driving differentiated execution of encryption strategies through semantic awareness, providing a new approach and method for image encryption.

[0071] This invention relates to The diffusion mechanism has been innovatively improved by expanding its core parameter n from a conventional fixed value (e.g., n=10) to a dynamic range n∈[8,12] controlled by the key, enabling the transformation matrix to adaptively adjust as the encryption process progresses. This mechanism embeds "..." into the encryption process. The "Dynamic Diffusion" module, combined with the key KEY1 generated by the SHA3-512 hash function, enables dynamic updates to the diffusion rules for each round, effectively mitigating security vulnerabilities caused by static parameters. For example, after image segmentation, dynamic n values ​​are enabled for key semantic regions. Diffusion to enhance resistance to attacks. This design retains... Based on the superior characteristics of transformation, the randomness, non-periodicity, and nonlinearity of chaotic systems are introduced, and significant technological breakthroughs are achieved through parameter dynamization.

[0072] This invention utilizes SHA3-512 hashing to extract image features and generate an initial key for a 2D-HNN chaotic system strongly correlated with the plaintext, achieving "one-time pad" encryption. Gradient segmentation is employed, with bit-level scrambling and dynamic Fibonacci transformation used to enhance encryption in high-semantic regions, while global scrambling is used in low-semantic regions, achieving a balance between security and efficiency. Multi-stage (bit, Fibonacci, block, diffusion) encryption is driven by a single chaotic system to ensure the consistency of the random source. This scheme effectively resists statistical and differential attacks and maintains decryption robustness under certain noise and masking conditions.

[0073] While maintaining its core innovative ideas, the technical solution proposed in this invention possesses excellent scalability and replaceability in its specific implementation modules, demonstrating the robustness and universality of the solution architecture. The semantic awareness module in the solution is not limited to the specific gradient calculation method described; any feature extraction algorithm capable of effectively identifying structures, contours, or salient regions in an image (such as Canny edge detection, frequency domain-based salient detection models, etc.) can serve as a functionally equivalent alternative to achieve intelligent segmentation of image content. Secondly, the power parameter of the dynamic Fibonacci transform in the solution takes values ​​from a preferred embodiment; in practice, it can be adjusted to other reasonable positive integer ranges based on a comprehensive consideration of security strength and computational efficiency. Finally, the chaotic system driving the entire encryption process requires the generation of pseudo-random sequences that meet cryptographic requirements. Therefore, in addition to the 2D-HNN system preferred in this invention, other systems with good chaotic characteristics (such as Logistic mapping, Henon mapping, or Chen systems, etc.) can also be integrated as equivalent random sources.

[0074] The above alternatives do not depart from the core protection scope of this invention, which is based on "semantic layering, dynamic obfuscation, and chaos-driven" principles. The embodiments described above are merely preferred embodiments of this invention and are not intended to limit the scope of the invention. Various modifications and improvements made to the technical solutions of this invention by those skilled in the art without departing from the spirit of the invention should fall within the protection scope defined by the claims.

Claims

1. A selective encryption method based on Fibonacci dynamic diffusion algorithm combined with semantic awareness, characterized in that, include: The original color image is acquired, and the importance of each region in the original color image is identified by a semantically perceptual-driven hierarchical encryption mechanism to obtain images of important regions and images of less important regions. The image of the important region is subjected to bit-level cross-scrambling to generate a three-dimensional tensor, and based on... The dynamic diffusion enhancement mechanism dynamically diffuses the important region image to obtain a dynamically diffused important region image; The dynamic diffusion enhancement mechanism is obtained by introducing position-related dynamic parameters into the original dynamic diffusion enhancement mechanism; Global chaotic block scrambling and chain-diffusion encryption are performed on the important region image and the secondary important region image after dynamic diffusion to obtain the final encrypted image.

2. The selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness fused according to claim 1, characterized in that, The method also includes: The original color image is subjected to dimensional constraint preprocessing and reshaped into a one-dimensional column vector; The one-dimensional column vector is input into the target hash function to generate a fixed-bit hash value, which is then converted into a target-bit unsigned integer numerical vector and mapped to a hash string composed of multiple hexadecimal characters. The hash string is converted into a decimal feature vector, which is used to calculate the initial state value, gain parameter, and weight matrix parameter of the 2D-HNN hyperchaotic system in order to construct a two-dimensional discrete chaotic model. The initial state is input into the two-dimensional discrete chaotic model, and after multiple pre-iterations to eliminate transient effects, iterative extraction continues. The component sequences are used to obtain the core one-dimensional sequence, and the PCHIP operator is used to perform nonlinear fitting on the core points to generate a full-length global chaotic sequence of the target length; the target length is the required total length estimated after multiple pre-iterations to eliminate transient effects. The full-length global chaotic sequence is divided into bit scrambling sequences. Diffusion sequence, global block scrambling sequence, and global diffusion parameters.

3. The selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness fused according to claim 1, characterized in that, Obtaining the important region image and the secondary important region image includes: The original color image is converted to a grayscale image, using... The operator performs a convolution operation on the grayscale image and calculates the horizontal and vertical gradients respectively. The combined gradient magnitude of each pixel is calculated using the horizontal and vertical gradients to construct a semantic gradient map. The semantic gradient map is flattened into a one-dimensional column vector and sorted in descending order of numerical value to obtain the corresponding original position index vector. The original color image is reshaped into a two-dimensional matrix, the row vectors of the two-dimensional matrix are rearranged according to the order of the original position index vectors, and then reshaped into the target three-dimensional tensor to obtain a semantically ordered image. The semantically ordered image is segmented according to the segmentation threshold to obtain the important region image and the secondary important region image.

4. The selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness fused according to claim 1, characterized in that, Generating the three-dimensional tensor includes: The important region image is reduced in dimension and reshaped into a one-dimensional pixel column vector. Bit extraction is then used to decompose each pixel in the one-dimensional pixel column vector into multiple binary bits to construct a global bit vector. The bit scrambling sequence is sorted in ascending order to generate a global position scrambling index vector, which is used to perform position mapping on the global bit vector to generate a scrambled bit vector. The scrambled bit vector is recombined into decimal pixel values ​​and summed according to binary weight rules to obtain a recombined pixel vector. The recombined pixel vector is then reverse-engineered into the three-dimensional tensor.

5. The selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness fused according to claim 1, characterized in that, The obtained image of the important region after dynamic diffusion includes: The three-dimensional tensor is converted into double-precision floating-point data to perform a dimension reshaping operation, which flattens the three-dimensional tensor and re-divides it into a pixel pair matrix. use The diffusion sequence maps chaotic sequence values ​​to a dynamic exponential sequence within an integer range, and confuses adjacent pixel vectors in the pixel pair matrix by performing a matrix power operation on the number of transformations corresponding to the dynamic exponential sequence, thereby generating a transformed pixel pair matrix. The transformed pixel pair matrix is ​​inversely reshaped to its original size and converted to an unsigned integer of the target bit to obtain the image of the important region after dynamic diffusion.

6. The selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness fusion according to claim 1, characterized in that, Performing the global chaotic block scrambling includes: The important region image and the secondary important region image after dynamic diffusion are tensor-stitched in the horizontal direction to reconstruct a full-size intermediate image, and then dimensionality-reduced and reshaped into a two-dimensional pixel matrix. The two-dimensional pixel matrix is ​​divided into multiple independent image blocks with the goal of treating multiple adjacent pixels as a single encryption unit, in order to construct an original pixel index sequence, and then the original pixel index sequence is reshaped into a block index matrix. The global block scrambling sequence is sorted in ascending order to obtain the block position permutation index vector. The block position permutation index vector is then used to perform a global permutation on the column vectors of the block index matrix to obtain the scrambled index matrix. The scrambled index matrix is ​​flattened into a one-dimensional index vector, which is used to extract pixel data from the two-dimensional pixel matrix, construct a scrambled two-dimensional matrix, and then reshape it into a three-dimensional image tensor of the target size.

7. The selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness fusion according to claim 6, characterized in that, Obtaining the final encrypted image includes: The three-dimensional image tensor is reduced and reshaped into a one-dimensional pixel input sequence, and a nonlinear quantization operation is performed using a global diffusion chaotic sequence to adapt to the target pixel position, thereby generating a diffusion key stream. The arithmetic-logic-displacement ternary hybrid diffusion model is used to break the linear mapping law through key-controlled dynamic displacement to forward diffuse the diffusion key stream and generate a one-dimensional ciphertext vector. The one-dimensional ciphertext vector is inversely reshaped into a three-dimensional image tensor and then converted into an unsigned integer of the target bit to obtain the final encrypted image.

8. The selective encryption method based on Fibonacci dynamic diffusion algorithm and semantic awareness fused according to claim 7, characterized in that, Generating the one-dimensional ciphertext vector includes: ； in, It is a one-dimensional ciphertext vector. This means using an 8-bit binary number Move left in a loop Bit, , For diffusion of key streams, The first pixel in the current input one-dimensional pixel sequence or the scrambled pixel sequence. pixel value, The ciphertext pixel values ​​obtained from the previous loop calculation. Modulo operation is used to truncate and restrict a value to a certain range. Within a single byte range.

9. A selective encryption system based on Fibonacci dynamic diffusion algorithm and semantic awareness, implemented according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to acquire raw color images; The semantic awareness module is used to identify the importance of each region in the original color image using a semantic awareness-driven hierarchical encryption mechanism, and to obtain images of important regions and images of less important regions. The dynamic diffusion module is used to perform bit-level cross-scrambling on the image of the important region, generate a three-dimensional tensor, and based on... The dynamic diffusion enhancement mechanism dynamically diffuses the important region image to obtain a dynamically diffused important region image; The dynamic diffusion enhancement mechanism is obtained by introducing position-related dynamic parameters into the original dynamic diffusion enhancement mechanism; The diffusion encryption module is used to perform global chaotic block scrambling and chain diffusion encryption on the important region image and the secondary important region image after dynamic diffusion to obtain the final encrypted image.

Images