Image encryption method and system based on bidirectional dynamic obfuscation and controllable noise diffusion

Abstract

The present invention discloses an image encryption method and system based on bidirectional dynamic obfuscation and controllable noise diffusion, belonging to the field of image encryption technology. The method includes obtaining a plaintext image, constructing and training a VortexCipherNet encryption network, and encrypting the plaintext image through the network based on a master key composed of multiple subkeys to obtain a ciphertext image. The VortexCipherNet encryption network is composed of a pixel grouping module, a first segmentation and exchange module, a noise injection module, a fully connected layer, a first waveform activation function, a one-dimensional convolutional layer, a second waveform activation function, a first mid-range dilated convolution, a third waveform activation function, a second mid-range dilated convolution, a Tanh activation function, and a second segmentation and exchange module connected in sequence. The present invention realizes efficient, secure, and dynamically adaptive image encryption through the design of bidirectional dynamic obfuscation, controllable noise diffusion, and a lightweight neural network.

Description

Technical Field

[0001] The present invention relates to the field of image encryption technology, and in particular to an image encryption method and system based on bidirectional dynamic obfuscation and controllable noise diffusion. Background Art

[0002] With the rapid development of digital imaging technology, the storage and transmission of image data are becoming increasingly frequent. However, the privacy and security of image data are becoming increasingly prominent. Traditional image encryption methods primarily rely on classic encryption algorithms, such as the Advanced Encryption Standard (AES). While these methods can ensure data security to a certain extent, their security is increasingly challenged by the increasing computing power and diverse attack methods. Furthermore, traditional encryption methods often require manual updates and adjustments, making them difficult to adapt to dynamically changing security requirements. In recent years, the application of neural networks in image processing has gradually attracted attention. Neural networks, with their powerful nonlinear fitting and adaptive capabilities, can provide new approaches to image encryption. However, most existing neural network-based image encryption methods suffer from high model complexity, low encryption efficiency, and difficulty in implementing dynamic adaptive encryption. Summary of the Invention

[0003] The present invention aims to provide an image encryption method and system based on bidirectional dynamic obfuscation and controllable noise diffusion, which realizes efficient, secure and dynamically adaptive image encryption through the design of bidirectional dynamic obfuscation, controllable noise diffusion and lightweight neural network.

[0004] The present invention is achieved through the following technical solutions.

[0005] An image encryption method based on bidirectional dynamic obfuscation and controllable noise diffusion, comprising the following steps:

[0006] Step 1: Get the plaintext image;

[0007] Step 2: Build and train the VortexCipherNet encryption network, which is composed of a pixel grouping module, a first split exchange module, a noise injection module, a fully connected layer, a first waveform activation function, a one-dimensional convolutional layer, a second waveform activation function, a first-level mid-range dilated convolution, a third waveform activation function, a second-level mid-range dilated convolution, a Tanh activation function, and a second split exchange module.

[0008] Step 3: Using the VortexCipherNet encryption network, the plaintext image is encrypted based on the master key K, where the master key K is composed of the first subkey K1, the second subkey K2, the third subkey K3, the fourth subkey K4, and the fifth subkey K5. The pixel grouping module groups the pixels of the plaintext image under the action of the first subkey K1. The first segmentation and exchange module segments the binary string under the action of the second subkey K2. The noise injection module performs noise addition under the control of the third subkey K3. The first waveform activation function, the second waveform activation function, and the third waveform activation function perform nonlinear obfuscation operations on the feature sequence under the participation of the fourth subkey K4 and the fifth subkey K5.

[0009] Step 4: Get the ciphertext image.

[0010] Further preferably, the process of encrypting the plaintext image using the VortexCipherNet encryption network is as follows:

[0011] The pixel grouping module groups the pixels of the plaintext image under the action of the first subkey K1, expands the grouped pixels bit by bit, and concatenates them with the first subkey K1 to obtain an n-bit long binary string;

[0012] The first segmentation exchange module, under the action of the second subkey K2, calculates the specific position of splitting the n-bit binary string, splits it and exchanges the two parts of the segmented data;

[0013] The noise injection module, under the control of the third subkey K3, uses the third subkey K3 as a seed to inject noise bound to the third subkey K3, generating a noise sequence of the same length as the input data;

[0014] The fully connected layer globally mixes the data after segmentation, exchange, and noise injection, maps the binary string after pixel grouping to the latent space, and obtains the feature sequence;

[0015] The waveform activation function performs nonlinear obfuscation operations on the feature sequence with the participation of the fourth subkey K4 and the fifth subkey K5; the first-level mid-range dilated convolution layer increases the receptive field of the convolution kernel, and the waveform activation function is combined with the transformation complexity to enhance the complexity;

[0016] The second-level mid-range dilated convolution layer downsamples the feature sequence again through dilated convolution and nonlinear waveform activation, fusing different channel and position information;

[0017] Finally, the feature sequence is nonlinearly transformed through the Tanh activation function so that the output features are in the range of [-1, 1].

[0018] More preferably, the fourth subkey K4 and the fifth subkey K5 are normalized as follows: .

[0019] More preferably, the specific steps of concatenating the first subkeys K1 to obtain an n-bit binary string are:

[0020] Step 1: Pixel grouping Expand the grouped pixels into a binary string L1;

[0021] Step 2: Concatenate with the 32-bit first subkey K1 to obtain a new binary string L2;

[0022] Step 3: Adjust the binary string L2 to n bits in length.

[0023] Further preferably, the first waveform activation function, the second waveform activation function and the third waveform activation function have the same structure and are defined as:

[0024] ,

[0025] in, is the output of the waveform activation function, w1 and w2 are both trainable activation parameters, and x is the input feature sequence.

[0026] The present invention also provides an image encryption system based on bidirectional dynamic obfuscation and controllable noise diffusion, comprising:

[0027] Data acquisition module, used to obtain plaintext images;

[0028] A VortexCipherNet encryption network is used to encrypt an input plaintext image. The VortexCipherNet encryption network is composed of a pixel grouping module, a first segmentation and exchange module, a noise injection module, a fully connected layer, a first waveform activation function, a one-dimensional convolutional layer, a second waveform activation function, a first-level medium-range dilated convolution, a third waveform activation function, a second-level medium-range dilated convolution, a Tanh activation function, and a second segmentation and exchange module, which are connected in sequence. The pixel grouping module groups pixels of the plaintext image under the action of a first subkey K1, the first segmentation and exchange module segments the binary string under the action of a second subkey K2, the noise injection module performs noise addition processing under the control of a third subkey K3, and the first waveform activation function, the second waveform activation function, and the third waveform activation function perform nonlinear obfuscation operations on the feature sequence with the participation of a fourth subkey K4 and a fifth subkey K5.

[0029] Beneficial effects of the present invention:

[0030] 1. Bidirectional dynamic obfuscation: Forward obfuscation is achieved by binding pixel grouping with a key (K1), and the waveform activation function with the fourth key K4 and the fifth subkey K5. Reverse obfuscation is achieved through a split exchange module and mid-range dilated convolution. The forward and reverse obfuscations form a closed-loop obfuscation, making it impossible for attackers to restore the plaintext image through local analysis. Dynamic exchange of positions and bidirectional exchange of neural network input and output enhances resistance to statistical analysis.

[0031] 2. Controllable noise diffusion: Noise diffusion is performed under the control of a key, bringing the information entropy of the ciphertext image close to the theoretical upper limit and enhancing the encryption effect.

[0032] 3. Lightweight neural network: The VortexCipherNet network has seven layers and 5K model parameters. It is a lightweight neural network suitable for real-time encryption needs.

[0033] 4. Dynamic adaptive encryption: Parameters such as split points, noise, and waveform are generated in real time by the key. Through adversarial training, the encryption strategy is adjusted in real time to adapt to dynamically changing security requirements.

[0034] 5. Coupling between keys and algorithms: The encryption process is deeply bound to the key. Even if the algorithm structure is public, different keys can produce completely different encryption behaviors, improving anti-reverse engineering capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Flow chart of the method of the present invention.

[0036] Figure 2 This is the master key composition diagram.

[0037] Figure 3 This is the VortexCipherNet network structure diagram.

[0038] Figure 4 This is the workflow diagram of the split exchange module.

[0039] Figure 5 Here are some examples of plaintext images.

[0040] Figure 6 This is an example of encryption and decryption effects. DETAILED DESCRIPTION

[0041] The present invention will be further described in detail below with reference to the accompanying drawings.

[0042] like Figure 1 As shown, this embodiment provides an image encryption method based on bidirectional dynamic obfuscation and controllable noise diffusion, the steps are as follows:

[0043] The image encryption method based on bidirectional dynamic obfuscation and controllable noise diffusion has the following steps:

[0044] Step 1. Obtain the plaintext image;

[0045] Step 2. Construct and train the VortexCipherNet encryption network;

[0046] Step 3. Use the VortexCipherNet encryption network to encrypt the plaintext image according to the master key K;

[0047] Step 4. Obtain the ciphertext image.

[0048] In this embodiment, the master key K is composed of the first sub-key K1, the second sub-key K2, the third sub-key K3, the fourth sub-key K4, and the fifth sub-key K5, as Figure 2 shown.

[0049] As Figure 3 shown, the VortexCipherNet encryption network is sequentially connected by a pixel grouping module, a first split and exchange module, a noise injection module, a fully connected layer, a first waveform activation function, a one-dimensional convolutional layer (Conv1D), a second waveform activation function, a first-level medium-range dilated convolution, a third waveform activation function, a second-level medium-range dilated convolution, a Tanh activation function, and a second split and exchange module. The specific encryption process is as follows:

[0050] The pixel grouping module groups the pixels of the plaintext image under the action of the first sub-key K1, expands the grouped pixels bit by bit, and connects them with the first sub-key K1 to obtain a binary string with a length of n bits;

[0051] The specific steps for connecting the first sub-key K1 to obtain a binary string with a length of n bits are:

[0052] The first step: Pixel grouping, which can be 2 pixels per group, or 4 or 8 pixels per group. And expand the grouped pixels into a binary string L1;

[0053] The second step: Connect with the 32-bit first sub-key K1 to obtain a new binary string L2. The connection method is determined by the 1st and 2nd bits of the first sub-key K1, as shown in the following table:

[0054]

[0055] The third step: Adjust the binary string L2 to a length of n bits. If L2 < n, perform the SHA256(K1) hash operation on the first sub-key K1, take the first n - L2 bits, and then splice the data into n bits. If L2 > n, truncate L2 to n bits; if L2 = n, directly output L2.

[0056] The first split exchange module, under the action of the second subkey K2, calculates the specific position of splitting the n-bit binary string (this step directly performs a modulo operation on n), splits it, and exchanges the two parts of the fragmented data;

[0057] The noise injection module, under the control of the third subkey K3, uses the third subkey K3 as a seed to inject noise bound to the third subkey K3, generating a noise sequence of the same length as the input data;

[0058] The fully connected layer globally mixes the data after segmentation, exchange and noise injection, and maps the binary string after pixel grouping to the latent space;

[0059] The waveform activation function performs nonlinear obfuscation operations on the data with the participation of the fourth subkey K4 and the fifth subkey K5. The first-level mid-range dilated convolution layer increases the receptive field of the convolution kernel and combines with the waveform activation function to enhance the complexity of the transformation. The fourth subkey K4 and the fifth subkey K5 are normalized as follows: ;

[0060] The second-level mid-range dilated convolution layer downsamples the sequence again through dilated convolution and nonlinear waveform activation, fusing different channel and position information;

[0061] Finally, the features are nonlinearly transformed through the Tanh activation function;

[0062] The pixel grouping module preprocesses the input data, converting high-dimensional pixel data into a binary sequence more suitable for network processing. This reduces the computational complexity of directly processing high-resolution images, thereby reducing network load. Using the first subkey K1, the pixel grouping module groups the pixels of the plaintext image. These pixels are then expanded bit by bit and concatenated with the first subkey K1 to produce an n-bit binary string. This design has the significant advantage that the influence of a single pixel bit can be diffused across multiple output bits through subsequent operations, achieving bit-level mixing. The inclusion of the first subkey K1 ensures that the same input produces completely different binary representations under different keys. After grouping, adjacent pixels may be assigned to different groups, disrupting the original spatial neighborhood relationships and destroying the statistical properties of the plaintext image.

[0063] The first segmentation exchange module calculates the specific position of the n-bit binary string under the action of the second subkey K2, and then splits it and further exchanges the two parts of the fragmented data. The process is as follows: Figure 4As shown, the input data is first checked for its shape, which can be 1D, 2D, or 3D. Then, operations are performed based on the data shape. If the data shape is 1D, data decompression is performed; if the data shape is 2D, it is processed directly; if the data shape is 3D, data contraction is performed. The split positions are obtained and the data is split into first and second slices based on the split positions. The first and second slices are swapped and connected, and the data shape is restored. The output is obtained through data contraction (1D), shape preservation (2D), and data decompression (3D). The split swap module scrambles the data positions, further disrupting the local correlation of the plaintext image, making it more difficult for attackers to infer the original structure through statistical analysis. Furthermore, it provides a more uniform input distribution for subsequent nonlinear transformations, preventing local feature leakage. The split swap module is used twice at the beginning and end of the model, creating bidirectional positional obfuscation and resisting position-based attacks.

[0064] The noise injection module injects noise under the control of the third subkey, K3. This randomized noise disrupts the determinism of the input data stream, causing the same input to produce different outputs across different runs. This also hides key patterns in the data, preventing reverse engineering of the model. Because the noise in the noise injection module is bound to the third subkey, K3, it can be considered a form of dynamic key expansion, further increasing the randomness of the encryption process. Even if an attacker inputs the same plaintext image multiple times, the output ciphertext image will be different due to the presence of noise. Furthermore, noise acts as an implicit regularization, preventing model overfitting, so this module also has a regularization effect.

[0065] In a fully connected layer, each input neuron is connected to all output neurons, and a weight matrix is used to linearly combine all dimensions of the input vector. Any change in any single bit of the input affects all output bits through the weight matrix, thus achieving the cryptographic diffusion goal. By changing the dimensions, the fully connected layer projects the input data into a new space of different dimensions, further hiding the original statistical characteristics and achieving global obfuscation. After the splitting, swapping, and noise injection of the input data, the fully connected layer globally mixes it, mapping the binary string of grouped pixels into the hidden space, further disrupting the original bit distribution. The obfuscation effect is further diffused through the local receptive fields of the mid-range dilated convolutions in the subsequent two layers.

[0066] The waveform activation function is defined as:

[0067] ,

[0068] in, is the output of the waveform activation function, w1 and w2 are both trainable activation parameters, and x is the input feature sequence.

[0069] The waveform activation function exhibits nonlinear obfuscation properties. The sine and cosine functions create highly nonlinear input and output relationships, resulting in a many-to-one mapping. Therefore, an attacker cannot uniquely determine the input from the output. Furthermore, the waveform activation function operates with the participation of the fourth and fifth subkeys, K4 and K5. During encryption, the values of K4 and K5 determine the specific form of the waveform transformation. Therefore, the waveform function varies with w1, w2, K4, and K5, resulting in completely different obfuscation effects depending on the key. Even if w1 and w2 are leaked, an attacker can obtain some plaintext-ciphertext pairs. However, due to the randomness of K4 and K5, the attacker cannot effectively infer the encryption results of other plaintext images. This also conforms to the Kerckhoffs design principle in cryptography.

[0070] The mid-range dilated convolutional layer uses dilation technology to increase the receptive field of the convolution kernel, allowing each output point to capture a wider range of the input. This allows features to interact and correlate over longer distances, which helps increase the complexity and nonlinearity of cryptographic features. The use of a first-level and second-level dilated convolutional layer increases the nonlinear correlation of features at different levels, making cryptographic features more difficult to analyze and predict. The dilation ratio of the first-level dilated convolutional layer is 2, meaning it effectively covers an input length of 5, or a process length of 5, while using the same number of parameters as a 3×1 convolution. This maintains computational efficiency while enhancing cross-region information mixing. Furthermore, considering the avalanche effect in cryptography, small changes such as single-bit flips in the input are propagated to multiple output locations through the dilated convolution process. This hopping propagation accelerates the avalanche effect, making the ciphertext image highly sensitive to changes in the plaintext image and key. The first-level dilated convolutional layer further enhances the transformation complexity through a nonlinear waveform activation function. The second-level mid-range dilated convolutional layer further enhances feature complexity through dilated convolution and nonlinear waveform activation. Simultaneously, it downsamples the sequence again, interleaving inputs from adjacent positions to the same output point while compressing the number of channels to 1. This process integrates information from different channels and positions, increasing the unpredictability of the ciphertext image. This multi-level design, equivalent to multiple rounds of iteration, increases feature complexity and makes cryptographic features more difficult to crack. The two-level mid-range dilated convolutional layer design also allows for feature extraction at different scales, enabling effective representation of cryptographic features at all scales. This increases the multi-scale nature of cryptographic features and further enhances cryptographic security. Finally, the features are nonlinearly transformed using the Tanh activation function, keeping the output features within the range [-1, 1], enhancing their boundedness and normalization.

[0071] Another embodiment of the present invention provides an image encryption system based on bidirectional dynamic obfuscation and controllable noise diffusion, including a data acquisition module and a VortexCipherNet encryption network. The data acquisition module is used to acquire plaintext images, and the VortexCipherNet encryption network encrypts the input plaintext images.

[0072] The experiments were conducted using the BOSSBase dataset, which consists of 10,240 grayscale images acquired by 7 digital cameras. Figure 5 Six of the plaintext images are given. Figure 6 for Figure 5 The encryption / decryption effect of plaintext image A, Figure 6In the figure, a is the plaintext image, b is the ciphertext image, c is the decrypted image, d is the horizontal correlation coefficient of the ciphertext image, e is the vertical correlation coefficient of the ciphertext image, and f is the diagonal correlation coefficient of the ciphertext image. Figure 5 The peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the encryption / decryption of the 6 plaintext images are shown in Table 1, and the entropy of the plaintext images and ciphertext images are shown in Table 2.

[0073] Table 1 Decrypted image PSNR / SSIM

[0074]

[0075] Table 2 Entropy of plaintext image and ciphertext image

[0076]

[0077] Experimental results demonstrate that this method achieves excellent encryption performance. The decrypted images achieved an average PSNR of 62.113 dB and an average SSIM of 0.9889. The average entropy of the ciphertext images was 7.9902, close to the theoretical value of 8 and an average increase of 7.71 percentage points over the plaintext image entropy.

[0078] The key length is the sum of the lengths of the four subkeys, i.e. 32+32+32+32=128, and the key space is up to 2 128 , meeting mainstream security requirements.

[0079] The above description merely represents preferred embodiments of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above disclosure to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention remain within the scope of protection of the present invention.

Claims

1. An image encryption method based on bidirectional dynamic obfuscation and controllable noise diffusion, characterized in that: The following steps are involved: Step 1: Get the plaintext image; Step 2: Build and train the VortexCipherNet encryption network, which is composed of a pixel grouping module, a first split exchange module, a noise injection module, a fully connected layer, a first waveform activation function, a one-dimensional convolutional layer, a second waveform activation function, a first-level mid-range dilated convolution, a third waveform activation function, a second-level mid-range dilated convolution, a Tanh activation function, and a second split exchange module. Step 3: Using the VortexCipherNet encryption network, the plaintext image is encrypted based on the master key K, where the master key K is composed of the first subkey K1, the second subkey K2, the third subkey K3, the fourth subkey K4, and the fifth subkey K5. The pixel grouping module groups the pixels of the plaintext image under the action of the first subkey K1. The first segmentation and exchange module segments the binary string under the action of the second subkey K2. The noise injection module performs noise addition under the control of the third subkey K3. The first waveform activation function, the second waveform activation function, and the third waveform activation function perform nonlinear obfuscation operations on the feature sequence under the participation of the fourth subkey K4 and the fifth subkey K5. Step 4: Get the ciphertext image.

2. The image encryption method according to claim 1, wherein: The process of encrypting a plaintext image using the VortexCipherNet encryption network is as follows: The pixel grouping module groups the pixels of the plaintext image under the action of the first subkey K1, expands the grouped pixels bit by bit, and concatenates them with the first subkey K1 to obtain an n-bit long binary string; The first segmentation exchange module, under the action of the second subkey K2, calculates the specific position of splitting the n-bit binary string, splits it and exchanges the two parts of the segmented data; The noise injection module, under the control of the third subkey K3, uses the third subkey K3 as a seed to inject noise bound to the third subkey K3, generating a noise sequence of the same length as the input data; The fully connected layer globally mixes the data after segmentation, exchange, and noise injection, maps the binary string after pixel grouping to the latent space, and obtains the feature sequence; The waveform activation function performs nonlinear obfuscation operations on the feature sequence with the participation of the fourth subkey K4 and the fifth subkey K5; the first-level mid-range dilated convolution layer increases the receptive field of the convolution kernel, and the waveform activation function is combined with the transformation complexity to enhance the complexity; The second-level mid-range dilated convolution layer downsamples the feature sequence again through dilated convolution and nonlinear waveform activation, fusing different channel and position information; Finally, the feature sequence is nonlinearly transformed through the Tanh activation function so that the output features are in the range of [-1, 1].

3. The image encryption method according to claim 2, wherein: The fourth subkey K4 and the fifth subkey K5 are normalized as follows: .

4. The image encryption method according to claim 2, wherein: The specific steps of concatenating the first subkey K1 to obtain an n-bit binary string are: Step 1: Pixel grouping Expand the grouped pixels into a binary string L1; Step 2: Concatenate with the 32-bit first subkey K1 to obtain a new binary string L2; Step 3: Adjust the binary string L2 to n bits in length.

5. The image encryption method according to claim 3, wherein: The first waveform activation function, the second waveform activation function, and the third waveform activation function have the same structure and are defined as: ， in, is the output of the waveform activation function, w1 and w2 are both trainable activation parameters, and x is the input feature sequence.

6. An image encryption system based on bidirectional dynamic obfuscation and controllable noise diffusion, characterized in that: include: Data acquisition module, used to obtain plaintext images; A VortexCipherNet encryption network is used to encrypt an input plaintext image. The VortexCipherNet encryption network is composed of a pixel grouping module, a first segmentation and exchange module, a noise injection module, a fully connected layer, a first waveform activation function, a one-dimensional convolutional layer, a second waveform activation function, a first-level medium-range dilated convolution, a third waveform activation function, a second-level medium-range dilated convolution, a Tanh activation function, and a second segmentation and exchange module, which are connected in sequence. The pixel grouping module groups pixels of the plaintext image under the action of a first subkey K1, the first segmentation and exchange module segments the binary string under the action of a second subkey K2, the noise injection module performs noise addition processing under the control of a third subkey K3, and the first waveform activation function, the second waveform activation function, and the third waveform activation function perform nonlinear obfuscation operations on the feature sequence with the participation of a fourth subkey K4 and a fifth subkey K5.

7. The image encryption system according to claim 6, wherein: The process of encrypting a plaintext image using the VortexCipherNet encryption network is as follows: The pixel grouping module groups the pixels of the plaintext image under the action of the first subkey K1, expands the grouped pixels bit by bit, and concatenates them with the first subkey K1 to obtain an n-bit long binary string; The first segmentation exchange module, under the action of the second subkey K2, calculates the specific position of splitting the n-bit binary string, splits it and exchanges the two parts of the segmented data; The noise injection module, under the control of the third subkey K3, uses the third subkey K3 as a seed to inject noise bound to the third subkey K3, generating a noise sequence of the same length as the input data; The fully connected layer globally mixes the data after segmentation, exchange, and noise injection, maps the binary string after pixel grouping to the latent space, and obtains the feature sequence; The waveform activation function performs nonlinear obfuscation operations on the feature sequence with the participation of the fourth subkey K4 and the fifth subkey K5; the first-level mid-range dilated convolution layer increases the receptive field of the convolution kernel, and the waveform activation function is combined with the transformation complexity to enhance the complexity; The second-level mid-range dilated convolution layer downsamples the feature sequence again through dilated convolution and nonlinear waveform activation, fusing different channel and position information; Finally, the feature sequence is nonlinearly transformed through the Tanh activation function so that the output features are in the range of [-1, 1].

Images