Compressed Sensing Reconstruction Based On Deep Learning And It’s Application In Image Encryption

With the rapid development of network communication,image data security is particularly important,and compressed sensing(CS)image encryption is a hot research field in image data security.Compressed sensing stores image data in the form of random measurements which can reduce the storage space.In the research of compressed sensing,reconstruction method is a hot topic.On the one hand,the traditional reconstruction methods based on iterative optimization have low running efficiency,and can not obtain ideal reconstruction results at low sampling ratio.On the other hand,the details of reconstruction image of the existing deep learning methods are fuzzy,and the visual performance can be improved.In addition,the existing compressed sensing encryption methods still have the problems of time-consuming reconstruction process,low sampling ratio and poor reconstruction quality.In order to solve these challenging problems,this paper studies and researches an image deep learning compressed sensing reconstruction and its encryption application based on deep learning and compressed sensing.The specific work contains the following aspects:(1)A framework named sampling and whole image denoising network based on generative adversarial network(GAN)for compressed sensing image reconstruction(SWDGAN)is proposed to solve the problems of poor reconstruction quality on low menasurement rates and running efficiency in traditional compressed sensing methods.We enhance the feature representation and remove the blocking artifacts by introducing a whole image dense residual denoising module(WDM)without affecting running efficiency.To improve the practicability and reconstruction quality of our algorithm,a fully connected network without bias is applied in sampling process.Moreover,we remove batch normalization(BN)layer to avoid the BN artifact on reconstruction image.The experimental results illustrate that our method outperforms the most advanced traditional methods and deep learning-based methods in terms of reconstruction quality and running time.(2)A robust compressed sensing image encryption algorithm based on generative countermeasure network,convolutional neural network(CNN)denoising network and chaotic system is proposed to solve the problems of slow encryption and decryption speed and poor reconstruction quality of compressed sensing image encryption algorithm in practical application.Firstly,the measurement of plain image is obtained by using sampling network.Secondly,the measurement results are scrambled by Logistic-Tent chaotic system to obtain the cipher image.During decryption,the cipher image is inversely scrambled to obtain the decrypted measurement,and the decrypted measurement is reconstructed into the decrypted reconstructed image through the reconstruction network.Finally,combined with CNN denoising network,the noise of decrypted reconstructed image is removed.Experiments show that our algorithm obtains higher reconstruction quality and running efficiency,and has better noise immunity and safety compared with the traditional compressed sensing encryption methods.(3)In order to solve the problem that the quality and texture details of gray and color image of existing compressed sensing reconstruction and encryption algorithms are poor,a deep image compressed sensing encryption network using multi-color space and texture feature(CSENMT)is proposed.The proposed multi-color space and texture network for image compressed sensing(MTNet)realize high visual performance by using the difference and inter-correlations of several color spaces and the texture information.In addition,adaptive permutation based on chaotic system and plain image(APCP)achieves a higher scrambling degree through the inter channel permutation operation compared with channel-by-channel permutation.The experimental results of gray image and color image show that this method has high advantages in reconstruction quality,visual expressiveness,efficiency and security.