Security Analysis Of A Class Of Chaotic Image Encryption Algorithm Based On Deep Learning

With the development of communication technologies and the popularity of mobile devices,people are becoming more accustomed to uploading multimedia information to the Internet such as digital images,videos and audios captured by mobile devices.However,due to the lack of dedicated encryption algorithms for multimedia information,most of images are stored in plaintext,increasing the risk of privacy issues and property problems caused by information disclosure.Therefore,proposing an image encryption algorithm with high security and good performance is the key to maintain the security of multimedia information.This urgent need has attracted the attention of many scholars and the image encryption algorithm based on chaotic system is a research hotspot.In this paper,we focus on the security of the encryption algorithm and analyze the weak point of the proposed algorithm.Based on chaotic system and deep learning,a hierarchical image encryption algorithm using chaotic system and Feistel network is proposed.Then the security of the algorithm is analyzed in several ways,including use deep convolutional network to simulate the drawbacks of encryption algorithms.The work of this paper can be summarized as below:(1)According to the characteristics of the image information distribution structure and considering the time used for the overall image encryption,during the encryption process,different bitplanes of gray images and color images are encrypted differently,the left four most significant bits(MSB)and the right four least significant bits(LSB)of each pixel are encrypted differently.The "incomplete Feistel structure" can encrypt different input positions use the different encryption intensity.The empirical analysis shows that using SHA2 to generate the hash value of the original image,then the initial value and parameters of the chaotic system are generated based on the hash value.The key streams generated by the chaotic sequences have the correlation with the original image which makes the encryption algorithm resistant to the selective plaintext attacks.The Feistel network increases security and the hierarchical structure shortens the encryption time,showing the proposed image encryption method is suitable for practical applications.(2)The deep convolutional network can grasp the common features of similar images and thus classify the images with high accuracy.Combined with this idea,we can use the deep convolutional network to classify the encrypted images,when the encryption process cannot completely cover the plain images information,the classification accuracy rate will be higher than the random prediction rate.The research shows that the hierarchical image encryption algorithm can completely cover the original image features,and the deep convolution network can not correctly predict the image classification with higher probability than random prediction.