Pseudo-ghost imaging encryption and decryption method based on Dense Net

Abstract

The present invention discloses a method for encrypting and decrypting pseudo-ghost imaging based on Dense Net, including encryption: (i) irradiating M random speckles onto a secret image to obtain a first bucket detection value; (ii) randomly arranging the first bucket detection value and inserting it into the random speckles of a camouflage image to obtain modulated random speckles, saving the randomly arranged first bucket detection value as key 1, and recording the position where the first bucket detection value is inserted as key 2; (iii) irradiating the modulated random speckles and the random speckles to form L speckle patterns onto the camouflage image to obtain a ciphertext; also including decryption: (i) using the ciphertext and the random speckles to recover the camouflage image; (ii) obtaining the randomly arranged bucket detection value by using key 2; (iii) obtaining the first bucket detection value by using key 1. (iv) Performing single-pixel imaging based on Dense Net on the first bucket detection value and the random speckles to recover the secret image. The present invention can effectively perform image encryption, improve the image encryption efficiency and facilitate information transmission.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and particularly relates to a method for encrypting and decrypting pseudo-ghost imaging based on Dense Net. Background Art

[0002] Ghost imaging, as a non-local imaging technology based on the intensity correlation fluctuation of light fields, has excellent characteristics such as high resolution, good anti-noise performance, and low light source limitation, and has received extensive attention in the field of computational imaging. Ghost imaging originated from the entangled two-photon imaging first demonstrated by Pittaman in 1995. In 2008, Shapiro et al. proposed computational ghost imaging, which does not require actual measurement and can directly obtain the reference light matrix through a computer, greatly simplifying the experimental setup of ghost imaging. As ghost imaging technology gradually matures, it also promotes its expansion into more application fields. For example, terahertz imaging proposed based on the advantages of ghost imaging in invisible light imaging has realized the reconstruction of a 3D image of an object through 4 bucket detectors, and remote sensing imaging proposed based on the good anti-interference performance of single pixels. What ghost imaging receives is the one-dimensional information of an object, which is convenient for transmission. Taking advantage of this feature, Clemente P et al. first applied ghost imaging to an optical encryption system in 2010. In 2012, Tanha et al. proposed a method for encrypting ghost imaging of grayscale images and color images. In 2013, Chen et al. proposed a method with a larger key storage space. In 2015, Zhao et al. proposed a high-performance encryption scheme for ghost imaging based on QR codes and compressive sensing. In 2016, Wu J et al. proposed a multi-image encryption method based on ghost imaging, improving the capacity of image encryption. Subsequently, in the field of ghost imaging encryption, encryption schemes for improving security performance have been continuously proposed.

[0003] As a new research field in the field of artificial intelligence, deep learning performs operations such as speech recognition and image processing by learning and analyzing the internal laws of data samples. In recent years, in the field of information optics, deep learning has also made certain breakthroughs in combination with scattering imaging, holographic imaging, etc. Of course, ghost imaging is no exception. In 2017, Lyu Meng et al. first proposed computational ghost imaging using deep learning. The traditional computational ghost imaging restored image and the corresponding original image were used to train a deep neural network model to improve the quality of image reconstruction. Y. He et al. proposed using the ghost imaging reconstruction map to train a convolutional neural network, realizing fast imaging under low sampling rate conditions. Subsequently, optimizing the ghost imaging effect through deep learning has gradually become one of the research hotspots. In addition to using neural networks to improve the signal-to-noise ratio of image reconstruction, some neural networks that can directly restore the original image using bucket detection values have also been successively proposed.

[0004] Traditional ghost imaging is an encryption method based on linear operations, which is prone to the risk of information theft. Moreover, due to the imaging characteristics of ghost imaging, most existing ghost imaging encryption methods require a large number of illumination matrices as keys, which is not convenient for transmission. Summary of the Invention

[0005] The object of the present invention is to provide a pseudo-ghost imaging encryption and decryption method based on Dense Net. The present invention can effectively encrypt images, improve the image encryption efficiency and facilitate information transmission at the same time.

[0006] To solve the above technical problems, the technical solution provided by the present invention is as follows: A pseudo-ghost imaging encryption and decryption method based on Dense Net, including an encryption process and a decryption process. The encryption process includes the following steps:

[0007] (i) Irradiate M random speckles as an illumination matrix onto a secret image to obtain a first bucket detection value;

[0008] (ii) After randomly arranging the first bucket detection values, insert them into the random speckles of the camouflage image to obtain modulated random speckles. Save the randomly arranged first bucket detection values as key 1, and record the positions where the first bucket detection values are inserted as key 2;

[0009] (iii) Combine the modulated random speckles in step (ii) and the random speckles in step (i) to form L speckle patterns, and irradiate the L speckle patterns onto the camouflage image to obtain a second bucket detection value, which is the ciphertext;

[0010] The decryption process includes the following steps:

[0011] (i) Use the ciphertext and random speckles to recover the camouflage image;

[0012] (ii) Use key 2 to determine the positions where the first bucket detection values are embedded, and then sum the modulated random speckles to obtain randomly arranged bucket detection values;

[0013] (iii) Recover the first bucket detection value with the correct sequence in the randomly arranged bucket detection values through key 1.

[0014] (iv) Perform single-pixel imaging based on Dense Net on the first bucket detection value and random speckles to recover the secret image.

[0015] The above pseudo-ghost imaging encryption and decryption method based on Dense Net, the overall network of Dense Net is based on Unet, with dense_block and transition_block as downsampling modules. After six layers of downsampling, it is connected to upsampling through one dense_block and dropout of 0.2. The upsampling part consists of one convolutionlayer and dense_block; the downsampling and upsampling are connected through skip connection layers and concatenation; the final output layer consists of a convolution layer and the activation function sigmoid.

[0016] In the aforementioned pseudo-ghost imaging encryption and decryption method based on Dense Net, in step (iii) of the encryption process, the number L of speckle patterns is greater than the number of random speckles in step (i).

[0017] In the aforementioned pseudo-ghost imaging encryption and decryption method based on Dense Net, in step (i) of the encryption process, the random speckles are represented as:

[0018]

[0019] where M represents the number of illumination patterns; N = a×a, which is the size of the illumination pattern;

[0020] The secret image is represented by the matrix S:

[0021] S = [S(1, 1) S(1, 2)…S(1, N)] T ;

[0022] where T represents the transpose of the matrix;

[0023] The first bucket detection value obtained by irradiating the random speckles on the secret image is represented by B, and the specific process is represented as:

[0024]

[0025] In the above manner, the first bucket detection value of the secret image is obtained.

[0026] Compared with the prior art, the present invention combines the principle of ghost imaging, hides the secret image information into the speckle pattern of the camouflage image, which cannot be cracked by traditional methods, can well deceive the eavesdropper, and improves the anti-theft performance of the encryption system; the present invention uses DenseNet as the decryption network, and even at a low sampling rate, the image can be well restored; the present invention further compresses the information transmission volume through the method of downsampling and neural network restoration, while enhancing the security of the system. In addition, the present invention makes the number L of the speckle patterns greater than the number M of the random speckles, so that the camouflage image can hide multiple secret images, and by setting different keys 2, it is ensured that the secret images are hidden in different positions of the random speckles of the camouflage image, achieving the effect of multi-image hiding. Brief Description of the Drawings

[0027] Figure 1 It is a flowchart of ghost imaging encryption and decryption.

[0028] Figure 2 It is a schematic diagram of ghost imaging.

[0029] Figure 3 It is a schematic diagram of the Dense Net network.

[0030] Figure 4 It is a reconstructed diagram of the correlation algorithm (CGI) and Dense Net-based ghost imaging (DNGI) at different sampling rates.

[0031] Figure 5 It is the simulation result of the present invention, (a) secret image; (b) pseudo-image; (c) ciphertext; (d) secret image restored by CGI; (e) secret image restored by DNGI; (f) secret image restored by CGI; (g) secret image restored by DNGI.

[0032] Figure 6 It is the simulation result of multi-image encryption.

[0033] Figure 7 It is the simulation result of high-pixel encryption, (a) secret image; (b) pseudo-image; (c) ciphertext; (d) secret image restored by CGI; (e) secret image restored by DNGI; (f) pseudo-image restored by CGI; (g) secret image restored by DNGI. Detailed Embodiment

[0034] The following further describes the present invention in conjunction with the embodiments and the drawings, but it is not used as the basis for limiting the present invention.

[0035] Embodiment: A pseudo-ghost imaging encryption and decryption method based on Dense Net, including an encryption process and a decryption process. The encryption process includes the following steps ( Figure 1 Encryption on the left side in

[0036] (i) Project M binary random speckles ( Figure 1 the illumination pattern in Figure 1 ) onto the secret image ( Figure 1 the secret image in

[0037]

[0038] ) to obtain the first bucket detection value. As

[0039] shown in

[0040] , the basic principle of single-pixel imaging is to project a series of illumination matrices without spatial resolution onto the object to be measured, and the obtained bucket detection value is subjected to a second-order correlation operation with the illumination matrix to restore the image of the object to be measured. The ghost imaging principle proposed in this embodiment is also implemented based on this principle. Among them, the random speckle is expressed as: T ;

[0041] where M represents the number of illumination patterns; N = a × a, which is the size of the illumination pattern;

[0042] The secret image is represented by the matrix S:

[0043]

[0044] S = [S(1, 1) S(1, 2) … S(1, N)]

[0045] where T represents the transpose of the matrix; Figure 1 The first bucket detection value obtained by irradiating the random speckle onto the secret image is represented by B, and the specific process is expressed as: Figure 1 where key2 in

[0046] By the above method, the first bucket detection value of the secret image is obtained. Figure 1 where the camouflaged image in

[0047] Among them, the number L of speckle patterns in step (iii) is greater than the number M of random speckles in step (i). In this way, the camouflage image can hide multiple secret images, and by setting different keys 2, it is ensured that the secret images are hidden in different positions of the random speckles of the camouflage image, achieving the effect of hiding multiple images.

[0048] The described decryption process includes the following steps ( Figure 1 the Decryption on the right in

[0049] (i) Recover the camouflage image using the ciphertext and random speckles;

[0050] (ii) Use key 2 to determine the position where the first bucket detection value is embedded, and then sum the modulated random speckles to obtain the scrambled bucket detection value;

[0051] (iii) Recover the first bucket detection value of the correct sequence in the scrambled bucket detection value through key 1.

[0052] (iv) Perform single-pixel imaging based on Dense Net on the first bucket detection value and random speckles to recover the secret image. In this embodiment, as Figure 3 shown, the overall network of the Dense Net is based on the Unet architecture, with dense_block and transition_block as downsampling modules. After six layers of downsampling, it is connected to the upsampling through one layer of dense_block and 0.2 dropout. The upsampling part consists of one layer of convolution layer and dense_block; the downsampling and upsampling are connected through skip connection layers and concatenation; the final output layer consists of a convolution layer and the activation function sigmoid.

[0053] Specifically, in this embodiment, the numerical recovery of the ghost imaging correlation algorithm (CGI) uses MATLAB2018. The decryption steps of DenseNet-based ghost imaging (DNGI) are completed based on the RTX4000 GPU computer experiment platform, the tensorflow1.12 framework, and the python3.6 development environment.

[0054] First, set the sampling rates (SR) to 10%, 5%, 1%, and 0.5% respectively. DNGI training uses the mnist dataset, where 10,000 images are selected as the training set and 1,000 as the test set. The training period is set to 50, the adam optimizer is used, and the learning rate is set to 0.02. The recovery effects of CGI and DNGI are compared, as Figure 4As shown, it can be seen that for CGI, when the SR is 10% and 5%, the outline of the graph can barely be seen, but most of the details are lost. When the SR is 1% and 0.5%, the image can no longer be distinguished. However, the DNGI adopted in the present invention can be restored even under extremely low sampling rates.

[0055] On the premise of ensuring the picture restoration effect, low sampling can improve the security of the encryption system. Here, the applicant sets the SR of the secret image to 1%. The sampling rate of the camouflage image is set to 10%. Numerical simulation is carried out according to the method of the present invention, and the results are as Figure 5 shown Figure 5 In (a) is the secret image to be encrypted, and (b) is the camouflage image used. The images are all 64*64. After encrypting the secret object with binary random speckles, the first bucket detection value is obtained and modulated. The obtained ciphertext is as Figure 5 (c). The ciphertext is transmitted to the receiver. The receiver decrypts the modulated random speckles to obtain (d), and decrypts with two keys to obtain the secret image (e). Then, after training with Dense Net, the restored secret image (f) and the camouflage image (g) are obtained.

[0056] When the sampling rates of the camouflage image and the secret image are the same, the camouflage image can hide multiple secret images. As Figure 6 shown, the applicant hides three secret images in the camouflage image. The sampling rate of the secret image is 1%, and the sampling rate of the pseudo-image is 10%. In this way, when the number of image pixels is the same, up to 10 secret images can be hidden.

[0057] As Figure 7 shown, (a) is a 128×128 secret image with a sampling rate of 1%, which is hidden into a pseudo-image (b) with a size of 64×64 and a sampling rate of 10%. The obtained ciphertext is (c). The secret image restored by CGI after decryption is (d), and the secret image restored by DNGI is (e). Similarly, the pseudo-image restored by CGI is (f), and the pseudo-image restored by DNGI is (g). It can be seen that the restoration effect of the image is still relatively good. Even when the number of pixels increases, the present invention can still efficiently and securely hide the image well.

[0058] In summary, the present invention combines the principle of ghost imaging to hide the secret image information into the speckle pattern of the camouflage image, which cannot be cracked by traditional methods, can well deceive the eavesdropper, and improves the anti-eavesdropping property of the encryption system; the present invention uses DenseNet as the decryption network, and even at a relatively low sampling rate, the image can be well restored; the present invention further compresses the information transmission volume through the method of downsampling and neural network restoration, while enhancing the security of the system. In addition, in the present invention, the number L of the speckle patterns is greater than the number M of the random speckles, so that the camouflage image can hide multiple secret images, and by setting different keys 2, it is ensured that the secret images are hidden in different positions of the random speckles of the camouflage image, achieving the effect of multi-image hiding.

Claims

1. Pseudo-ghost imaging encryption and decryption method based on Dense Net, characterized in that: It includes an encryption process and a decryption process. The encryption process includes the following steps: (i) Use M random speckles as the illumination matrix to irradiate the secret image to obtain the first bucket detection value; (ii) After scrambling the first bucket detection value, insert it into the random speckles of the camouflage image to obtain the modulated random speckles. Save the scrambled first bucket detection value as key 1, and record the position where the first bucket detection value is inserted as key 2; (iii) Combine the modulated random speckles in step (ii) and the random speckles in step (i) to form L speckle patterns, and irradiate the L speckle patterns on the camouflage image to obtain the second bucket detection value, which is the ciphertext; The decryption process includes the following steps: (i) Use the ciphertext and the speckle pattern to restore the camouflage image; (ii) Use key 2 to determine the position where the first bucket detection value is embedded, and then sum the modulated random speckles to obtain the scrambled bucket detection value; (iii) Restore the first bucket detection value in the correct sequence in the scrambled bucket detection value through key 1; (iv) Perform single-pixel imaging based on Dense Net on the first bucket detection value and the random speckles to restore the secret image; The overall network of the Dense Net is based on the Unet architecture, with the dense_block and transition_block as the downsampling modules. After six layers of downsampling, it is connected to the upsampling through one dense_block and 0.2 dropout. The upsampling part consists of one convolutionlayer and dense_block; the downsampling and upsampling are connected through skip connection layers and concatenation; the final output layer consists of a convolution layer and the activation function sigmoid; In step (i) of the encryption process, the random speckles are represented as: where M represents the number of illumination patterns; N = a×a, which is the size of the illumination pattern; The secret image is represented by the matrix S: S = [S(1, 1) s(1, 2) … s(1, N)] T ; where T represents the transpose of the matrix; The first bucket detection value obtained by irradiating the random speckles on the secret image is represented by B, and the specific process is represented as: In the above way, the first bucket detection value of the secret image is obtained.

2. The pseudo-ghost imaging encryption and decryption method based on Dense Net according to claim 1, characterized in that: In step (iii) of the encryption process, the number L of the speckle patterns is greater than the number M of the random speckles in step (i).

Images