Remote sensing image zero watermarking method based on multichannel pulse coupling neural network

Abstract

The invention discloses a remote sensing image zero watermarking method based on a multichannel pulse coupling neural network. According to the method, firstly, all wave bands of a remote sensing image are input into a multi-channel pulse coupling neural network model for parallel processing, and the model is optimized and excited by adopting a multi-scale morphological gradient and a differential box dimension, so that the accuracy and robustness of feature extraction are improved. And then, the extracted features are encrypted by using asymmetric Tent mapping, and the watermark image and the encrypted feature image are scrambled according to generalized Arnold transformation, so that the security of the algorithm is improved. And finally, performing XOR operation on the scrambled watermark image and the feature image to generate a zero watermark, and storing the zero watermark in an interstellar file system (IPFS). The method has relatively high robustness on non-geometric watermark attacks and geometric watermark attacks of common remote sensing images, and provides reliable guarantee for security management and copyright protection of the remote sensing images.

Description

Technical Field

[0001] The present invention belongs to the field of spatial data security, and particularly relates to a zero-watermark method for remote sensing images based on a multi-channel pulse-coupled neural network. Background Art

[0002] With the rapid development of earth observation technology, the amount of remote sensing image data has increased exponentially. In particular, high-resolution remote sensing images (HRRS) have shown great application potential in key fields such as environmental monitoring, urban planning, disaster assessment, and agricultural management. However, the unique value and wide application of HRRS data have also highlighted the importance of copyright protection. Although traditional digital watermarking techniques can effectively identify the copyright ownership, the embedding process inevitably modifies the original image data, affecting the data accuracy. Especially in high-precision demand scenarios such as map making and geoscience analysis, the problem of data distortion is particularly prominent. Therefore, the zero-watermark technology for remote sensing images has emerged. It uses a non-embedding method to construct a copyright information identifier by extracting the inherent features of the image without changing the original image data, which not only ensures the integrity and accuracy of HRRS data but also realizes the effective protection of copyright, meeting the high-fidelity requirements for data in the field of remote sensing applications. Despite the great potential shown by the zero-watermark technology for remote sensing images, it still faces challenges. For example, how to efficiently extract stable and unique image features to construct a robust copyright identifier, and how to quickly and accurately verify the authenticity of copyright information in copyright disputes. These problems still need to be further studied and solved.

[0003] By summarizing the existing technologies, it is found that there are at least the following technical problems:

[0004] 1) Remote sensing images integrate multiple features such as spectrum, shape, texture, and spatial relationship. However, most current methods are limited to traditional artificial feature extraction means, with limited applicability and insufficient feature representation ability. It is difficult to fully cope with the complex and diverse characteristics of remote sensing images, and it is also vulnerable to directional attacks, resulting in insufficient robustness of the zero-watermark method and affecting the effect and accuracy of remote sensing image analysis and application.

[0005] 2) When facing remote sensing images of the same geographical area but with different acquisition times and sensors, existing zero-watermark technologies may generate similar results. This limitation reduces their effectiveness in identifying and verifying the copyright ownership of remote sensing images in the same region, making it difficult to ensure the uniqueness and non-replicability of copyright information, and thus affecting the effectiveness and accuracy of copyright protection. Summary of the Invention

[0006] The present invention proposes a zero-watermark method for remote sensing images based on a multi-channel pulse-coupled neural network, aiming to solve the problems of insufficient robustness and limited accuracy existing in current digital watermarking technologies.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A zero-watermark method for remote sensing images based on a multi-channel pulse-coupled neural network, comprising:

[0009] S1: Construct a multi-channel pulse-coupled neural network (MCPCNN) model containing multiple parallel processing channels. Each channel has an independent pulse-coupled neural network structure, and the information between channels is weighted according to their respective contribution degrees;

[0010] S2: Calculate the multi-scale morphological gradient of each band of the remote sensing image and use it as the input of each channel of the model, so as to realize the optimization of the external stimulus of the MCPCNN;

[0011] S3: Adaptively express the connection strength of the MCPCNN according to the differential box dimension idea and perform normalization processing through the Sigmoid function;

[0012] S4: Perform n iterations on the optimized MCPCNN to obtain n pulse images, then calculate the entropy value of each pulse image, and arrange them in time series to obtain the feature vector of the remote sensing image;

[0013] S5: Perform NSST transformation on the R, G, and B bands of the remote sensing image respectively, input the corresponding three low-frequency components into the MCPCNN to calculate the feature vector of the remote sensing image, and then perform quantization processing on the feature vector;

[0014] S6: Generate a chaotic sequence according to the asymmetric Tent map using the key 1, and generate an encrypted feature image by using displacement operation on the quantized feature vector and the chaotic sequence;

[0015] S7: Scramble the watermark image and the encrypted feature image respectively using the key 2 and the key 3 according to the generalized Arnorld transform, and then perform exclusive OR calculation on the two to generate a zero watermark;

[0016] S8: After the copyright owner sends the key 1, key 2, key 3, zero watermark and authentication information to IPFS, IPFS verifies the received information and generates a certificate;

[0017] S9: After the verifier downloads the hash function and the certificate from IPFS, calculate a new certificate through hash according to the key 1, key 2, key 3, zero watermark and authentication information, and compare it with the downloaded certificate to determine whether the information at hand is true and complete;

[0018] S10: Similar to the algorithm of the zero-watermark construction part, the verifier first extracts the features of the remote sensing image to be verified through MCPCNN, then obtains the encrypted feature image according to the asymmetric Tent mapping, and finally scrambles the encrypted feature image using the generalized Arnold transform;

[0019] S11: The verifier performs an exclusive OR operation on the obtained scrambled feature image and the zero-watermark to obtain a scrambled version of the retrieved watermark, and then obtains the detected watermark through the inverse generalized Arnold transform.

[0020] Compared with the prior art, the present invention has the following beneficial technical effects:

[0021] Construct an MCPCNN containing multiple parallel processing channels to improve the applicability of the model in zero-watermark applications and the robustness to invariant features. Optimize the MCPCNN using multi-scale morphological gradients to enhance the detail expression ability and spatial correlation of the image. Stimulate the connection strength of the MCPCNN according to the differential box dimension to better measure the surface texture features of the image. In addition, encrypt and scramble the feature image according to the asymmetric Tent mapping and the generalized Arnold transform to improve the security of the method. Description of the Drawings

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0023] Figure 1 It is a schematic diagram of the proposed multi-channel pulse coupled neural network;

[0024] Figure 2 It is a flowchart of zero-watermark construction;

[0025] Figure 3 It is a flowchart of zero-watermark verification. Detailed Embodiments

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0027] An embodiment of the present invention discloses a remote sensing image zero-watermark method based on a multi-channel pulse coupled neural network, which includes three parts: multi-channel pulse coupled neural network model construction, zero-watermark construction, and zero-watermark verification:

[0028] The process of constructing the multi-channel pulse coupled neural network model is as follows:

[0029] Step 1: The standard PCNN model has a complex structure and numerous parameter settings. To improve the applicability of the model in zero-watermark applications and the robustness to invariant features, a multi-channel PCNN (MCPCNN) based on multi-scale morphological gradient and fractal dimension excitation is proposed, which mainly consists of a receptive field, a modulation field, and a pulse generator. The architecture of its model is shown in the appendix Figure 1 Its mathematical expression is

[0030]

[0031] In the formula, U ij (n), Y ij (n), and E ij (n) are the internal activity term, output result, and dynamic threshold of the neuron at position (i, j) at the nth iteration, respectively; is the external stimulus at position (i, j) in the pth channel, that is, the gray value of the corresponding position of the image; f = 0.2 and g = 0.9 are attenuation coefficients, h = 20 is the amplitude coefficient, γ = 1 is a constant, W ilkl = [0.707, 1, 0.707; 1, 0, 1; 0.707, 1, 0.707], C p is the weight of the th channel, and β p is the connection strength of the pth channel by differential box dimension excitation.

[0032] Step 2: When the MCPCNN model receives an external stimulus, its feedback input and the link input ∑ kl W ijkl Y kl (n - 1) are both adjusted by the connection strength coefficient β p , and then the internal activity U ij (n) is generated. Among them, the weighting coefficient C p is calculated according to the average gray value μ 1 , μ 2 , …, μ p , … μ m , where m is the number of channels, and its expression is

[0033]

[0034] Subsequently, this internal activity is compared with the dynamic threshold E in the previous iteration cycle ij (n - 1) to determine whether the neuron generates a pulse output Y ij (n). If the neuron is activated, i.e., Y ij (n) = 1, its dynamic threshold E ij (n) will suddenly increase by an amplitude of h. Conversely, if the neuron is not activated, i.e., Y ij (n) = 0, then the dynamic threshold E ij (n) will gradually decay in each iteration cycle according to the coefficient g until the internal activity U ij (n) is greater than the dynamic threshold E ij (n - 1) of the previous iteration cycle, at which point the neuron will be activated again. Subsequently, the generated pulse output signal Y ij (n) will be fed back to the input end of the neuron and take effect in the next iteration. By iterative solution, a series of pulse images can be generated.

[0035] Step 3: Select the MSMG of the p-th channel as the external stimulus of the model to obtain higher normalization values and prediction rates, so as to better quantify the detailed information of the image and enhance the spatial correlation of the image. Its expression is

[0036]

[0037] In the formula, represents the pixel at the position (i, j) of the p-th channel. SE is the basic structure element, the radius of SE is set to 3, the radius of the structure element of scale t is set to 2t + 1, and N is the total number of scales. and ⊙ respectively represent the morphological dilation and erosion operators. w t = 1 / (2t + 1) represents the weight of the t-th scale gradient.

[0038] Step 4: According to the idea of differential box dimension, the connection strength β of MCPCNN p is adaptively represented and normalized through the Sigmoid function. Its expression is

[0039]

[0040] In the formula, and respectively represent the maximum and minimum gray levels within a window size of 3×3 centered on the pixel at the position (i, j) of the th channel, and η = 0.01 is a constant.

[0041] Step 5: For the remote sensing image When the MCPCNN iterates n times, n pulse images will be generated, namely Y(1), Y(2), …, Y(n). The invariant features of the image are obtained from the time series generated by the pulse images. This time series records the activation order of neurons and reflects the pulse synchronization characteristics. Compared with the method of directly calculating the number of activated neurons in each pulse image to obtain the time series, the entropy of the pulse image is more suitable for measuring the information contained in the pulse image. The calculation method of the entropy value H(P) of the pulse image is

[0042] H(P) = -P1log2(P1) - P0log2(P0) (Equation 5)

[0043] where P1 and P0 are the probabilities of 1 and 0 appearing in the pulse image. Arranging the entropy values of the pulse images in order can obtain the feature vector of the remote sensing image N W is the total number of iterations of the MCPCNN, that is, the dimension of the feature vector.

[0044] The process of constructing the zero-watermark is as follows:

[0045] Step 6: The flow chart of zero-watermark construction is shown in the appendix Figure 2 . First, perform NSST transformation on the R, G, and B channels of the remote sensing image respectively, and input the corresponding three low-frequency components into the MCPCNN to calculate the feature vector v of the remote sensing image. Then, quantize the feature vector v to obtain which can be expressed as

[0046]

[0047] where mean(·) represents the calculation of the average value.

[0048] Step 7: Obtain the encrypted feature image according to the asymmetric Tent mapping.

[0049] The first step: Define the key key1 = {l = N W , σ = 0.45, x0 = 0.5}, where l represents the number of iterations, σ is the control parameter, and x0 represents the initial value.

[0050] The second step: Generate a chaotic sequence using key1 according to the asymmetric Tent mapping.

[0051] The third step: Encrypt the quantized feature vector bv and the chaotic sequence using a displacement operation to generate the feature image I B , and the number of displacement bits is determined according to the asymmetric Tent mapping.

[0052] Step 8: Scramble the watermark image and the encrypted feature image.

[0053] Step 1: Define the secret key key2 = {a = 2, b = 3, tt = 10} and the secret key key3 = {a = 1, b = 2, tt = 5}, where a and b are scrambling parameters and tt is the number of iterations.

[0054] Step 2: Use key2 and key3 respectively to scramble the encrypted feature image I B and the watermark image I W through the generalized Arnold transform, thereby improving the security of the proposed algorithm. The scrambling algorithm generates the scrambled versions of the feature image I BS and the watermark I WS . The expression of the generalized Arnold transform is

[0055]

[0056] where (i′, j′) represents the pixel position after the transformation of (i, j).

[0057] Step 3: Generate the zero watermark ZW = XOR(I BS , I WS ) according to the XOR operation.

[0058] Step 9: The copyright owner sends the secret key 1, secret key 2, secret key 3, and the zero watermark ZW, as well as the authentication information, to IPFS. IPFS verifies the received information and generates a certificate Cert = hash(ZW||key1||key2||key3||ID Owner ). hash(·) is a hash function; || represents the concatenation of marked data.

[0059] The zero watermark verification process is as follows:

[0060] Step 10: The flowchart of the zero watermark verification can be seen in Appendix Figure 3 . First, the verifier downloads the hash function hash(·) and the certificate Cert from IPFS. Then, the verifier performs a hash calculation based on the secret keys (key1, key2, key3), ZW, and the identity identifier ID Owner to obtain a new certificate, that is, Cert' = hash(ZW||key1||key2||key3||ID Owner ). Finally, compare Cert′ and Cert. If they are the same, proceed to the next step. Otherwise, it indicates that the information held by the verifier is untrue or incomplete.

[0061] Step 11: The same as the algorithm in the zero watermark construction part. First, extract the feature bv′ of the remote sensing image to be verified through MCPCNN, and then obtain the encrypted feature image I′ according to the asymmetric Tent mapping B, finally, the encrypted feature image is scrambled using the generalized Arnold transform to obtain I′ BS .

[0062] Step 12: Detect the watermark image.

[0063] The first step: Use the exclusive OR operation to obtain the scrambled version I′ of the retrieved watermark WS = XOR(I′ BS , ZW).

[0064] The second step: The verifier obtains the watermark I′ through the generalized Arnold inverse transform W . The expression is

[0065]

[0066] The third step: Compare I′ W and I W . If the two are the same, it is considered to pass the verification, otherwise it is considered not to pass the verification.

[0067] The present invention has the following advantages compared with the prior art:

[0068] First, by constructing the MCPCNN model containing multiple parallel processing channels, the applicability of the model in the application of zero watermark for remote sensing images is significantly improved. In addition, the multi-scale morphological gradient is used to optimize the MCPCNN, further enhancing the detail expression ability and spatial correlation of the image. According to the differential box dimension to stimulate the link strength of the MCPCNN, the texture features of the image surface are better measured, and the robustness of the model to invariant features is improved.

[0069] Second, by introducing technologies such as asymmetric Tent mapping and generalized Arnold transform, not only the security of the feature image is improved, but also the entire zero watermark method has stronger defense capabilities against different attack means. In addition, the present invention not only protects the privacy of the watermark information, but also ensures the integrity and authenticity of the watermark information.

[0070] Third, by combining the IPFS decentralized storage and zero watermark technology, not only the traditional storage bottleneck is solved, the efficient data distribution and fast access are realized, but also the uniqueness and immutability of the data are ensured. At the same time, this technology adds a hidden and robust copyright identifier to the data, provides strong copyright protection, and has high flexibility and scalability.

[0071] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0072] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing image zero watermark method based on a multi-channel pulse coupled neural network, including three parts: multi-channel pulse coupled neural network model construction, zero watermark construction and zero watermark verification: The steps for building a multi-channel pulse coupled neural network model are as follows: S1: Construct a multi-channel pulse coupled neural network (MCPCNN) model with multiple parallel processing channels. Each channel has an independent pulse coupled neural network structure, and the information between channels is weighted according to their respective contributions. S2: Calculate the multi-scale morphological gradient of each band of the remote sensing image and use it as the input of each channel of the model to optimize the external stimulation of MCPCNN; S3: Adaptively represent the link strength of MCPCNN based on the difference box dimension idea and normalize it using the Sigmoid function; S4: Execute iterations on the optimized MCPCNN to obtain n pulse images, then calculate the entropy value of each pulse image, and arrange them in time sequence to obtain the feature vector of the remote sensing image; The steps to construct a zero watermark are as follows: S5: Perform NSST transformation on the three bands of R, G, and B of the remote sensing image respectively, and input the corresponding three low-frequency components into MCPCNN to calculate the feature vector of the remote sensing image, and then quantize the feature vector; S6: Generate a chaotic sequence using key 1 according to the asymmetric Tent mapping, and use a shift operation to generate an encrypted feature image using the quantized feature vector and the chaotic sequence; S7: according to the generalized Arnorld transform, the watermark image and the encrypted feature image are scrambled using key 2 and key 3 respectively, and then the two are XORed to generate a zero watermark; S8: After the copyright owner sends key 1, key 2, key 3, zero watermark and authentication information to IPFS, IPFS verifies the received information and generates a certificate; The steps of zero watermark verification are as follows: S9: After the verifier downloads the hash function and certificate from IPFS, it calculates a new certificate based on key 1, key 2, key 3, zero watermark and authentication information, and compares it with the downloaded certificate to determine whether the information it has is true and complete; S10: The algorithm is the same as the zero watermark construction part. The verifier first extracts the features of the remote sensing image to be verified through MCPCNN, then obtains the encrypted feature image based on the asymmetric Tent mapping, and finally scrambles the encrypted feature image using the generalized Arnorld transform. S11: The verifier performs an XOR operation on the obtained scrambled feature image and the zero watermark to obtain the scrambled version of the retrieval watermark, and then obtains the detected watermark through the generalized inverse Arnorld transform.

Images