A high-security image encryption and decryption system and method combining an algorithm layer and a physical layer

Abstract

The application discloses a high-security image encryption and decryption system and method combining an algorithm layer and a physical layer, and relates to the technical field of image encryption, optical communication and deep learning cross. The system comprises a transmitting end and a receiving end. The transmitting end divides and randomly shuffles an original image, and generates an extended ciphertext block by using an AES-256 module encryption with Reed-Solomon error correction code fusion. Pixel gray values are converted into binary sequences, a vortex light mode sequence is generated and loaded into a spatial light modulator. The vortex light beam is converted into a random speckle pattern by a dynamic scattering device for transmission. The receiving end uses a pre-trained deep residual network to identify and demodulate the speckle image frame by frame, and then performs Reed-Solomon error correction decoding, AES-256 decryption and inverse shuffling operations in sequence to restore the original image. The application significantly improves the anti-brute-force cracking and anti-channel error code capability by deeply integrating digital encryption, physical layer vortex light encoding and dynamic scattering, and realizes high-precision and low-delay receiving processing by combining deep learning demodulation.

Description

Technical Field

[0001] This invention relates to the fields of image encryption, optical communication and deep learning, and in particular to a high-security image encryption and decryption system and method that combines algorithm layer and physical layer. Background Technology

[0002] Traditional image encryption primarily relies on digital cryptographic algorithms, such as the Advanced Encryption Standard (AES) and chaotic maps, which offer high security in terms of computational complexity. However, relying solely on algorithm-level encryption has inherent drawbacks: on the one hand, the security assumptions based on mathematical problems in traditional digital encryption algorithms may be challenged when facing new computing paradigms such as quantum computing; on the other hand, when the encrypted ciphertext is transmitted in fiber optic or free-space optical communication links in the form of conventional intensity modulation or phase modulation, attackers can use methods such as optical beam splitting and coherent detection to intercept the complete ciphertext and then obtain the key through offline brute-force attacks or side-channel attacks. Therefore, relying solely on algorithm-level encryption is insufficient to address scenarios where the physical layer signal can be completely eavesdropped.

[0003] Optical encryption technology has become a research hotspot in the field of secure communication in recent years due to its high parallelism, multidimensional modulation capability, and physical non-cloning properties. Among them, vortex optical coding technology based on orbital angular momentum (OAM) is particularly prominent: the vortex beam carries a helical phase factor exp(ilθ), and its topological charge l can theoretically take any integer value, thus providing a high-dimensional key space. However, pure vortex optical encryption systems face severe challenges in practical optical communication links.

[0004] In existing technologies, digital encryption and physical layer modulation are independent and lack deep integration, making it difficult to form a collaborative defense mechanism. Atmospheric turbulence or scattering media can cause crosstalk and phase distortion in vortex optical modes, leading to a sharp increase in the bit error rate of traditional interferometric demodulation methods, and lacking an active error correction mechanism for channel errors. Existing systems rely on computationally intensive and poorly generalized traditional phase recovery algorithms (such as the Gerchberg-Saxton iteration) for decryption, which cannot meet real-time requirements and have poor robustness to dynamic scattering environments. Most existing vortex optical encryption schemes only use the OAM mode itself as the information carrier, lacking deep integration with high-strength digital encryption algorithms and insufficient security redundancy.

[0005] In summary, existing technologies suffer from several prominent problems, including a disconnect between the algorithm and physical layers, insufficient robustness in decryption under optical link disturbances, a lack of proactive error correction mechanisms for ciphertext transmission errors, and a lack of efficient demodulation methods incorporating deep learning. To address these issues, this invention proposes a multi-level high-security encryption and decryption method that integrates strong digital encryption, vortex optical physical coding, dynamic scattering encryption, and intelligent demodulation. Summary of the Invention

[0006] The purpose of this invention is to provide a high-security image encryption and decryption system and method that combines algorithmic and physical layers. It combines digital algorithm layer encryption consisting of image block scrambling, AES-256 block encryption, and error correction code extension with physical layer encryption consisting of vortex light mode mapping and dynamic scattering medium speckle to form a multi-level encryption structure, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a high-security image encryption and decryption system combining algorithm and physical layers, comprising a transmitter and a receiver, wherein the transmitter includes:

[0008] The image segmentation and scrambling module is used to divide the original image into multiple N×N pixel image blocks and scramble the image blocks in a random order using a scrambling key.

[0009] The AES encryption and error correction module is used to perform AES-256 encryption on each scrambled image block, and to perform Reed-Solomon error correction encoding on the encrypted ciphertext to generate a check symbol and append it to the end of the ciphertext data, forming a ciphertext block that is expanded to M×M pixels, where M>N.

[0010] The vortex light coding module is used to convert the gray value of each pixel in the ciphertext block into a binary sequence, map each 4-bit binary combination into a vortex light topological charge mode, generate a vortex light mode sequence, load the mode sequence into the spatial light modulator in sequence to generate the corresponding vortex beam, and after passing through the dynamic scattering medium, the vortex beam is converted into a random speckle pattern and emitted into the channel.

[0011] The receiving end includes:

[0012] The deep learning demodulation module is used to acquire speckle patterns, output the probability distribution of vortex light patterns using a pre-trained deep residual network, and take the class with the highest probability as the demodulation result, thereby decoding and reconstructing M×M ciphertext blocks.

[0013] The AES decryption and error correction module is used to perform Reed-Solomon error correction decoding on the M×M ciphertext block to obtain the N×N ciphertext block, and then use the AES-256 key to perform inverse transformation decryption to obtain the N×N plaintext block.

[0014] The image restoration and reconstruction module is used to reverse scramble all plaintext blocks according to the scrambling key and then stitch them together to restore the original image.

[0015] Preferably, the dynamic scattering medium includes a scattering medium and a random vibration module, wherein the vibration frequency of the random vibration module is configured to be lower than the camera acquisition frequency, thereby keeping the speckle statistical characteristics quasi-static within an image block transmission cycle.

[0016] Preferably, the deep residual network is based on ResNet-50 architecture, with 16 output nodes in the last fully connected layer, and uses the Softmax activation function.

[0017] Preferably, the error correction capability of the Reed-Solomon error correction code is set to correct at least t symbol errors, and the number of verification symbols is not less than 2t.

[0018] Preferably, in the AES encryption and error correction module, the check symbols generated by Reed-Solomon encoding are sequentially appended to the end of the ciphertext data byte sequence after AES encryption, forming an extended byte sequence in which information symbols and check symbols are arranged continuously. The extended byte sequence is rearranged in row-major order into a two-dimensional ciphertext block of M×M pixels, where M² = N² + 2t, and t is the symbol error correction capability of the Reed-Solomon code.

[0019] Preferably, in the vortex optical coding module, the mapping relationship between the 16 4-bit binary combinations and the 16 different vortex optical topological charges is either fixed and preset or dynamically changed according to the second key.

[0020] Another technical problem this invention aims to solve is to provide a high-security image encryption and decryption method that combines algorithmic and physical layers, comprising the following steps:

[0021] The transmitter executes:

[0022] S1: Divide the original image into N×N pixel patches, padding with zeros if necessary, and scramble the patch order using a random scrambling key;

[0023] S2: Perform AES-256 encryption on each scrambled tile to obtain ciphertext data; perform Reed-Solomon encoding on the ciphertext data to generate an encoded block containing information symbols and check symbols, and rearrange it into an M×M pixel ciphertext tile;

[0024] S3: Convert the grayscale value of each pixel in the encrypted image block into 8-bit binary, and map each group of 4 bits into a vortex light topological charge mode to generate a vortex light mode sequence.

[0025] S4: Generate a time-varying random speckle pattern from the vortex light pattern sequence through a dynamic scattering medium and send it;

[0026] Receiver executes:

[0027] S5: The receiver acquires speckle patterns frame by frame after transmission through the channel at the same frame rate as the vortex light mode switching frequency of the transmitter, ensuring that each frame of speckle pattern corresponds one-to-one with each frame of vortex light mode of the transmitter in time. The acquired speckle pattern is input into a pre-trained deep convolutional neural network, which outputs the probability distribution of 16 types of vortex light modes, and the class with the highest probability is taken as the demodulation result.

[0028] S6: Decode the demodulated pattern sequence into 4-bit binary according to the mapping rule, merge them into 8-bit pixel values, and reconstruct the M×M ciphertext block;

[0029] S7: Perform Reed-Solomon error correction decoding on the M×M ciphertext block to obtain an N×N ciphertext block;

[0030] S8: Perform AES-256 decryption on the N×N ciphertext block using the AES-256 key to obtain the N×N plaintext block;

[0031] S9: Reverse scramble all plaintext blocks according to the scrambling key and then stitch them together to restore the original image.

[0032] Preferably, in step S5, the collected speckle pattern is normalized to 224×224 pixels and then input into a deep residual network; the deep residual network outputs a 16-dimensional probability vector, and after the 16-dimensional probability vector is normalized by the Softmax activation function, the category index corresponding to the maximum probability value is taken as the vortex topological charge pattern recognition result of the current frame; the recognition results of two adjacent frames are combined in chronological order and restored to an 8-bit grayscale value of a pixel.

[0033] Preferably, in step S7, the Reed-Solomon error correction decoding uses the Berlekamp-Massey algorithm to calculate the error location polynomial, determines the error symbol location through Chien search, and then uses the Forney algorithm to calculate the error magnitude to complete the correction of the error symbol.

[0034] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a high-security image encryption and decryption system and method that combines algorithmic and physical layers. It combines digital algorithm layer encryption, consisting of image block scrambling, AES-256 block encryption, and error correction code extension, with physical layer encryption, consisting of vortex light mode mapping and dynamic scattering medium speckle, to form a multi-level encryption structure. Attackers need to simultaneously crack the AES-256 key, scrambling rules, vortex light mode mapping rules, and the dynamic speckle characteristics of the scattering medium. Compared with schemes that only use AES or only use vortex light modulation, the multi-level encryption architecture of this invention greatly increases the difficulty of common cryptanalysis techniques such as known-plaintext attacks and chosen-ciphertext attacks. Due to the physical non-cloning property of dynamic scattering, it is difficult to recover the plaintext even if the light signal is intercepted.

[0035] This invention introduces an error correction code extension mechanism into the AES-256 encryption framework, expanding the ciphertext block size from N×N to M×M. When the vortex light signal causes partial mode crosstalk and demodulation errors after passing through the scattering medium, the receiver can use the error correction code to detect and correct errors before AES decryption. With proper configuration of RS code parameters, a certain number of pixel symbol errors can be completely corrected, effectively avoiding the problem of a single bit error causing the entire block to fail to decrypt.

[0036] The present invention employs a pre-trained deep residual network as the core for speckle image recognition and demodulation, which can directly map the corresponding vortex light pattern sequence from complex dynamic speckle to the end without relying on traditional interferometry or iterative phase recovery algorithms. It avoids the strict requirements of physical models on channel state information. On ordinary GPUs, the demodulation time per frame can be as low as about 30ms, which is about two orders of magnitude faster than the traditional Gerchberg-Saxton iterative method, meeting the requirements of real-time transmission.

[0037] This invention maps binary ciphertext sequences onto multiple vortex optical modes with different topological charges. Each optical symbol can carry 4 bits of information. Compared with traditional binary intensity modulation, the physical layer coding efficiency of this invention is multiplied. At the same time, the combination of high-dimensional modes constitutes a huge key space, further increasing the difficulty of brute-force cracking. Attached Figure Description

[0038] Figure 1 This is a block diagram of the overall architecture of the high-security image encryption and decryption system combining the algorithm layer and the physical layer of the present invention;

[0039] Figure 2 This is a schematic diagram of the internal structure of the AES-256 encryption algorithm used in the method of this invention.

[0040] Figure 3 This is a schematic diagram of the experimental device for vortex light generation, transmission and speckle acquisition in Embodiment 1 of the present invention;

[0041] Figure 4 The images show the phase diagram and intensity image of the vortex light in Embodiment 1 of the present invention.

[0042] Figure 5 This is a schematic diagram of the ResNet-50 network model used for demodulation in Embodiment 2 of the present invention;

[0043] Figure 6 This is a graph showing the loss function and accuracy curves during the training process of the ResNet-50 network in Embodiment 2 of the present invention.

[0044] Figure 7 This is a diagram illustrating the specific implementation process of the three-level image encryption in Embodiment 3 of the present invention;

[0045] Figure 8 This is a one-to-one mapping encoding diagram between the 16 vortex light modes and the binary numbers of image pixels in Embodiment 3 of the present invention;

[0046] Figure 9 This is a diagram illustrating the recovery process of the three images being decrypted step by step in an embodiment of the present invention;

[0047] Figure 10 This is a diagram showing the security test results of simulating various decryption methods in Embodiment 4 of the present invention. Detailed Implementation

[0048] 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.

[0049] Example 1: Construction of a Vortex Light Generation and Dynamic Scattering Device

[0050] Please see Figures 3-4 ,like Figure 3 As shown, the optical experimental setup for generating vortex light is constructed in the following order: laser, attenuator, polarizer, lens L1, lens L2, SLM, lens L3, aperture, lens L4, dynamic scattering medium, lens L5, and CCD camera. Both the SLM and CCD camera are connected to a computer. A stable Gaussian beam is generated using a 532nm semiconductor laser. This Gaussian beam is passed through an attenuator and polarizer to adjust its intensity and polarization state, and then expanded by a beam expander system consisting of lenses L1 and L2 with focal lengths of 50mm and 200mm, respectively. The expanded Gaussian beam is then perpendicularly incident on a pure phase spatial light modulator (SLM). A vortex phase diagram with a preset topological charge mode is applied to the SLM, as shown below. Figure 4 As shown, the Gaussian beam is modulated to generate vortex beams with different topological charge modes. The emitted vortex beams then enter a 4f spatial filtering system consisting of lens L3, a spatial filter stop, and lens L4 to isolate and extract the +1st order diffracted vortex beams, thus obtaining pure vortex beams.

[0051] The filtered, purified vortex beam is passed through a self-built dynamic scattering medium, composed of a 600-mesh scattering medium and a random vibrator, which dynamically and randomly scatters and modulates the pure vortex beam, thus forming a random speckle pattern in space. Lens L5, with a focal length of 30mm, is used to collect the speckle image, which is ultimately acquired by a CCD camera. The vibration frequency of the random vibration module is set to 5Hz, while the vortex beam mode switching frequency is 60Hz. This configuration ensures that the speckle statistical characteristics of the scattering medium remain quasi-static throughout a complete image patch transmission cycle, thereby meeting the channel stability requirements for subsequent deep learning demodulation.

[0052] Example 2: Dataset Construction and Deep Learning Demodulation Network Training

[0053] Please see Figures 5-6 For 16 different vortex topological charge modes, corresponding to 16 4-bit binary combinations, 10,000 speckle sample images were collected for each mode under the aforementioned dynamic scattering medium environment, resulting in a dataset containing a total of 160,000 images. The dataset was divided into training, testing, and validation sets in an 8:1:1 ratio, with the training set containing 128,000 images, and the testing and validation sets each containing 16,000 images.

[0054] After uniformly resizing the speckle image to 224×224 pixels, input as follows: Figure 5 The ResNet-50 network shown is based on the ResNet-50 architecture, with the number of output nodes in the last fully connected layer modified to 16, and a Softmax activation function used to output the probability distributions of 16 vortex light modes. The network is trained using supervised learning, with the input being a speckle image and the label being the actual OAM mode category (one-hot encoded) modulated by the transmitter. The training parameters are as follows: initial learning rate set to 0.001, optimizer set to Adam, loss function set to classification cross-entropy, training epochs set to 30, and batch size set to 64.

[0055] like Figure 6 As shown, the loss function tends to converge stably after about the 21st epoch, and the final classification accuracy on the test set reaches 99.78%. This result proves that the proposed deep residual network can achieve high-precision vortex beam pattern recognition in complex dynamic speckle environments, laying the foundation for the lossless reconstruction of subsequent ciphertext patches.

[0056] Example 3: Specific Implementation of Image Encryption and Decryption

[0057] Please see Figures 1-2 A high-security image encryption and decryption system combining algorithm and physical layers, comprising a transmitter and a receiver, wherein the transmitter includes:

[0058] The image segmentation and scrambling module is used to divide the original image into multiple N×N pixel image blocks and scramble the image blocks in a random order using a scrambling key.

[0059] The AES encryption and error correction module is used to perform AES-256 encryption on each scrambled image block, and to perform Reed-Solomon error correction encoding on the encrypted ciphertext to generate a check symbol and append it to the end of the ciphertext data, forming a ciphertext block that is expanded to M×M pixels, where M>N.

[0060] The vortex light coding module is used to convert the gray value of each pixel in the ciphertext block into a binary sequence, map each 4-bit binary combination into a vortex light topological charge mode, generate a vortex light mode sequence, load the mode sequence into the spatial light modulator in sequence to generate the corresponding vortex beam, and after passing through the dynamic scattering medium, the vortex beam is converted into a random speckle pattern and emitted into the channel.

[0061] The receiving end includes:

[0062] The deep learning demodulation module is used to acquire speckle patterns, output the probability distribution of vortex light patterns using a pre-trained deep residual network, and take the class with the highest probability as the demodulation result, thereby decoding and reconstructing M×M ciphertext blocks.

[0063] The AES decryption and error correction module is used to perform Reed-Solomon error correction decoding on the M×M ciphertext block to obtain the N×N ciphertext block, and then use the AES-256 key to perform inverse transformation decryption to obtain the N×N plaintext block.

[0064] The image restoration and reconstruction module is used to reverse scramble all plaintext blocks according to the scrambling key and then stitch them together to restore the original image.

[0065] Please see Figures 7-9 This embodiment uses an actual image as an example to fully demonstrate the entire processing flow of the transmitting and receiving ends.

[0066] Step 1: Image preprocessing at the transmitting end

[0067] like Figure 7 As shown, this embodiment selects a 256-level grayscale image with a size of 110×100 pixels as the original image. Since its size cannot be divided evenly by the 50×50 pixel block size, zero-value pixels are added to the right and bottom edges of the image to expand it into a 150×150 pixel extended image. Subsequently, the extended image is divided into nine 50×50 pixel sub-blocks, Block 1 to Block 9, and the arrangement order of the blocks is shuffled according to random scrambling rules, such as moving Block 1 to the end, to destroy spatial correlation and complete the first level of encryption.

[0068] Step 2: Digital Layer AES and Error Correction

[0069] Each scrambled 50×50 pixel patch is input into the AES-256 encryption module. AES-256 uses a 128-bit block length to encrypt the patch data block by block. To combat pixel symbol errors that may be introduced during subsequent optical transmission and deep learning demodulation, the system cascades a Reed-Solomon error correction code mechanism within the AES-256 framework.

[0070] In this embodiment, the RS code parameters are chosen as (n, k) = (72, 36), where the number of check symbols is 36. Specifically, for each 50×50 pixel AES-encrypted ciphertext data, RS(3672, 3600) encoding is performed to generate a 72-byte check symbol, which is appended to the end of the ciphertext, and then rearranged into a 60×60 pixel ciphertext patch. This expansion process expands the original 2500 bytes of information to 3600 bytes, with an expansion factor M / N = 60 / 50 = 1.2. After the combined processing of AES-256 encryption and RS encoding, each patch achieves a second level of encryption.

[0071] Step 3: Optical Modulation and Physical Layer

[0072] like Figure 7 As shown, the grayscale value (0-255) of each pixel in the 60×60 pixel ciphertext block is determined according to... Figure 8 The encoding table shown is converted into an 8-bit binary sequence, and the 8 bits are split into two groups of 4 bits and 4 bits. According to the preset one-to-one mapping relationship between 16 binary combinations and 16 vortex topological charge modes, each pixel corresponds to two frames of vortex beams. The entire encrypted image block is encoded into 7200 vortex beam mode sequences.

[0073] The SLM is controlled to sequentially modulate and transmit the corresponding vortex beams at a switching frequency of 60Hz. When the beams pass through the self-built scattering medium, they are subjected to dynamic physical scattering, forming a random speckle pattern that changes over time and is transmitted to the free space channel. The scattering medium consists of a 220-mesh scattering medium and a random vibrator with a vibration frequency of 5Hz. This process completes the third level of encryption.

[0074] Step 4: Demodulation and Reconstruction of Receiver Network

[0075] like Figure 9As shown, the receiving CCD camera captures the speckle pattern transmitted through the channel frame by frame at the same frame rate as the transmitting end. The speckle image is normalized to 224×224 pixels, and each frame of speckle image is input into the trained ResNet-50 demodulation network. The network outputs a 16-dimensional probability vector, and the category corresponding to the highest probability is taken as the demodulation result of that frame, thus obtaining 7200 pattern recognition results.

[0076] Due to the influence of dynamic scattering environment and camera noise, a small number of errors may occur during the classification and recognition process. Under the signal-to-noise ratio of 15dB in this embodiment, the actual measurement shows that the number of demodulated error pixels does not exceed 36. The demodulated pattern sequence is decoded into 4-bit binary according to the mapping rule, and two adjacent groups of 4-bit codes are merged into 8-bit pixel values ​​to reconstruct a 60×60 pixel degraded ciphertext patch with potential errors.

[0077] Subsequently, the ciphertext block is fed into the AES decryption and error correction module. First, Reed-Solomon error correction decoding is performed on the ciphertext block data: using the accompanying 72-byte checksum, erroneous symbols are located and corrected according to the Berlekamp-Massey algorithm of the RS code. Since the number of erroneous symbols does not exceed the error correction capability of the RS code t=36, all errors are completely corrected, and the original 50×50 pixel ciphertext data is extracted without loss. Then, using the correct AES-256 key, AES-256 decryption is performed on the ciphertext data to obtain a 50×50 pixel plaintext block.

[0078] Finally, based on the tile scrambling key recorded during the encryption phase, a reverse scrambling operation was performed on all nine decrypted plaintext sub-tiles, restoring them to their original spatial positions. After stitching, the zero-value boundaries filled due to size expansion were removed, restoring a 110×100 pixel grayscale image that is completely consistent with the original image. Verification showed that the reconstructed image is pixel-by-pixel consistent with the original image, with an error of 0.

[0079] Example 4: Security Testing

[0080] Please see Figure 10 To verify the security of the multi-level encryption system of this invention, decryption simulation tests were conducted on three typical attack scenarios. The tests used the same original image and the same encryption parameters as in Example 3. The results are as follows: Figure 10 As shown.

[0081] (a) Incorrect encoding rules only

[0082] Attack Scenario: Suppose an attacker obtains the correct AES-256 key and tile reassembly order through some means, but is unaware of the mapping rule between the vortex light pattern and 4-bit binary data, i.e., uses an incorrect mapping table to decode the ResNet-50 output. Since the mapping rule constitutes an independent security layer, the ciphertext pixel sequence decoded by the attacker is completely different from the actual ciphertext. Even if subsequent decryption is performed with the correct AES-256 key, the input is already erroneous data, and the final decryption result presents a random noise distribution unrelated to the original plaintext, such as... Figure 10 As shown, the peak signal-to-noise ratio approaches 0 dB, and the structural similarity index (SSIM) is below 0.1. This result indicates that the OAM coding rules themselves constitute an indispensable key.

[0083] (b) AES key error only

[0084] Attack Scenario: Assume the attacker correctly identifies the OAM encoding rules and the tile reassembly order, but uses an incorrect AES-256 key for decryption. The AES-256 algorithm has a key space of 2^256, making exhaustive search infeasible with current computing power. When performing AES-256 decryption on correct ciphertext tiles using an arbitrary incorrect key, the output is a statistically uniformly distributed pseudo-random byte sequence. For example... Figure 10 As shown, the decryption results for each image patch all exhibit uniform noise, with a correlation coefficient of less than 0.01 with the original image, making it impossible to extract any image information. This test verifies that the AES-256 module provides a high-strength cryptographic barrier at the digital level.

[0085] (c) Recombination rule error only

[0086] Attack Scenario: Assume the attacker possesses both the correct OAM encoding rules and the AES-256 key, but is unaware of the scrambling order of the image tiles. In this embodiment, the original image is divided into 9 sub-tiles, with a total permutation count of 9! = 362,880. Combined with the spatial uncertainty introduced by edge zero-value padding, the actual equivalent key space is even larger. When the attacker reassembles the correctly decrypted sub-tiles in an incorrect reassembly order, while the texture details within each sub-tile are correctly preserved, the overall spatial layout of the image is completely chaotic, failing to form a recognizable global structure, such as... Figure 10 As shown in the figure. This result indicates that tile scrambling constitutes a valid third key.

[0087] Based on the test results of the three attack scenarios mentioned above, the multi-level fusion optical encryption system proposed in this invention can still effectively prevent plaintext recovery even if the key of a single level is leaked, demonstrating high security redundancy and anti-attack capability.

[0088] Example 5: Real-time Testing

[0089] To evaluate the real-time processing capability of the system of the present invention, a time test was conducted on the receiving end processing flow on a computer equipped with an Intel Core i7-14700 CPU and an NVIDIA RTX 4070 GPU.

[0090] The test sample consisted of 1000 consecutively received speckle pattern sequences.

[0091] The measurement results are as follows:

[0092] Single-frame speckle image preprocessing, normalized to 224×224: average 0.2ms;

[0093] Single-frame ResNet-50 forward inference: average 0.87ms;

[0094] RS error correction decoding and AES-256 decryption, single tile: average 7.7ms;

[0095] Reverse scrambling and image stitching: average 23.3ms.

[0096] The average total delay from receiving a single frame to the final plaintext output is approximately 32.1 ms. The processing speed at the receiver can match the data rate at the transmitter, meeting the requirements for real-time image transmission. In contrast, the traditional Gerchberg-Saxton iterative phase retrieval method has a demodulation time of approximately 3 seconds per frame, which cannot achieve real-time processing.

[0097] Example 6: Channel Error Resistance Test

[0098] To quantitatively evaluate the effectiveness of the built-in RS error correction mechanism, the signal-to-noise ratio (SNR) at the receiver was varied in a dynamic scattering channel. By adjusting the attenuator or increasing ambient light noise, the final bit error rate (BER) of the system and its ability to recover the image without loss were measured under different SNR conditions. The BER was expressed as the pixel error rate. The test results are shown in the table below:

[0099] Table 1. Summary of Anti-channel Performance Test Results

[0100] Signal-to-noise ratio (dB) Average number of error pixels after demodulation Number of error pixels after RS ​​correction Image restoration results 20 3 0 lossless 15 28 0 lossless 12 45 9 Local damage 10 78 42 Unable to recover

[0101] When the signal-to-noise ratio (SNR) is ≥15dB, the number of demodulation error pixels is ≤36, which is within the error correction capability of the RS code, enabling complete lossless recovery. When the SNR drops to 12dB, the number of errors exceeds 36, and residual errors remain after RS ​​error correction, causing some image blocks to fail to decrypt. This test demonstrates that the RS error correction mechanism set in this invention improves the system's error tolerance under typical operating conditions by approximately 10dB. Compared to the situation without error correction, where approximately 28 error pixels directly lead to AES decryption failure at 15dB, this significantly enhances the system's robustness to channel disturbances.

[0102] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-security image encryption and decryption system combining algorithmic and physical layers, characterized in that, It includes a transmitter and a receiver, wherein the transmitter includes: The image segmentation and scrambling module is used to divide the original image into multiple N×N pixel image blocks and scramble the image blocks in a random order using a scrambling key. The AES encryption and error correction module is used to perform AES-256 encryption on each scrambled image block, and to perform Reed-Solomon error correction encoding on the encrypted ciphertext to generate a check symbol and append it to the end of the ciphertext data, forming a ciphertext block that is expanded to M×M pixels, where M>N. The vortex light coding module is used to convert the gray value of each pixel in the ciphertext block into a binary sequence, map each 4-bit binary combination into a vortex light topological charge mode, generate a vortex light mode sequence, load the mode sequence into the spatial light modulator in sequence to generate the corresponding vortex beam, and after passing through the dynamic scattering medium, the vortex beam is converted into a random speckle pattern and emitted into the channel. The receiving end includes: The deep learning demodulation module is used to acquire speckle patterns, output the probability distribution of vortex light patterns using a pre-trained deep residual network, and take the class with the highest probability as the demodulation result, thereby decoding and reconstructing M×M ciphertext blocks. The AES decryption and error correction module is used to perform Reed-Solomon error correction decoding on the M×M ciphertext block to obtain the N×N ciphertext block, and then use the AES-256 key to perform inverse transformation decryption to obtain the N×N plaintext block. The image restoration and reconstruction module is used to reverse scramble all plaintext blocks according to the scrambling key and then stitch them together to restore the original image.

2. The high-security image encryption and decryption system combining algorithm layer and physical layer according to claim 1, characterized in that, The dynamic scattering medium includes a scattering medium and a random vibration module. The vibration frequency of the random vibration module is configured to be lower than the camera acquisition frequency, so that the speckle statistical characteristics remain quasi-static within one image block transmission cycle.

3. The high-security image encryption and decryption system combining algorithm layer and physical layer according to claim 1, characterized in that, The deep residual network is based on ResNet-50 architecture, with 16 output nodes in the last fully connected layer and employing the Softmax activation function.

4. The high-security image encryption and decryption system combining algorithm layer and physical layer according to claim 1, characterized in that, The error correction capability of the Reed-Solomon error correction code is set to correct at least t symbol errors, and the number of verification symbols is not less than 2t.

5. A high-security image encryption and decryption system combining algorithm and physical layers according to claim 4, characterized in that, In the AES encryption and error correction module, the check symbols generated by Reed-Solomon encoding are sequentially appended to the end of the ciphertext data byte sequence after AES encryption, forming an extended byte sequence in which information symbols and check symbols are arranged continuously. The extended byte sequence is rearranged in row-major order into a two-dimensional ciphertext block of M×M pixels, where M² = N² + 2t, and t is the symbol error correction capability of the Reed-Solomon code.

6. A high-security image encryption and decryption system combining algorithm and physical layers according to claim 1, characterized in that, In the vortex optical coding module, the mapping relationship between 16 four-bit binary combinations and 16 different vortex optical topological charges is either fixed and preset or dynamically changed according to the second key.

7. A high-security image encryption and decryption method combining algorithm layer and physical layer as described in any one of claims 1-6, characterized in that, Includes the following steps: The transmitter executes: S1: Divide the original image into N×N pixel patches, padding with zeros if necessary, and scramble the patch order using a random scrambling key; S2: Perform AES-256 encryption on each scrambled tile to obtain ciphertext data; perform Reed-Solomon encoding on the ciphertext data to generate an encoded block containing information symbols and check symbols, and rearrange it into an M×M pixel ciphertext tile; S3: Convert the grayscale value of each pixel in the encrypted image block into 8-bit binary, and map each group of 4 bits into a vortex light topological charge mode to generate a vortex light mode sequence. S4: Generate a time-varying random speckle pattern from the vortex light pattern sequence through a dynamic scattering medium and send it; Receiver executes: S5: The receiver acquires speckle patterns frame by frame after transmission through the channel at the same frame rate as the vortex light mode switching frequency of the transmitter, ensuring that each frame of speckle pattern corresponds one-to-one with each frame of vortex light mode of the transmitter in time. The acquired speckle pattern is input into a pre-trained deep convolutional neural network, which outputs the probability distribution of 16 types of vortex light modes, and the class with the highest probability is taken as the demodulation result. S6: Decode the demodulated pattern sequence into 4-bit binary according to the mapping rule, merge them into 8-bit pixel values, and reconstruct the M×M ciphertext block; S7: Perform Reed-Solomon error correction decoding on the M×M ciphertext block to obtain an N×N ciphertext block; S8: Perform AES-256 decryption on the N×N ciphertext block using the AES-256 key to obtain the N×N plaintext block; S9: Reverse scramble all plaintext blocks according to the scrambling key and then stitch them together to restore the original image.

8. The high-security image encryption and decryption method combining algorithm layer and physical layer according to claim 7, characterized in that, In step S5, the collected speckle pattern is normalized to 224×224 pixels and then input into a deep residual network. The deep residual network outputs a 16-dimensional probability vector. After the 16-dimensional probability vector is normalized by the Softmax activation function, the category index corresponding to the maximum probability value is taken as the vortex topological charge pattern recognition result of the current frame. The recognition results of two adjacent frames are combined in chronological order and restored to an 8-bit grayscale value of a pixel.

9. A high-security image encryption and decryption method combining algorithm layer and physical layer according to claim 7, characterized in that, In step S7, the Reed-Solomon error correction decoding uses the Berlekamp-Massey algorithm to calculate the error location polynomial, determines the error symbol location through Chien search, and then uses the Forney algorithm to calculate the error magnitude, thus completing the correction of the error symbol.

Images