Image Encryption Based On Variational Auto-encoder Genertive Models

With the development of computer networks and information technologies,multimedia communication has become an important means for people to exchange information.As the most important form of information expression in multimedia communication,digital images have obviously become the mainstream of information expression.However,digital images bring people convenience,as well as many problems that security transmit and store.For example,some image information involve state secrets,trade secrets,or personal privacy.But the information can be maliciously attacked,tampered,illegally copied and propagated through the Internet.it is also very important to quickly transfer information in a limited bandwidth.In particular,the emergence of cloud storage,it can not only store a large number of information data such as files,videos and images,but also provide a large enough online space to store shared data.In this case,the requirements for real-time and security of information transmission are even stricter.Therefore,it is of great theoretical and practical significance to reduce the bandwidth of the image and ensure the security of the image encryption.Image encryption,also known as digital image encryption,mainly protects the image by using different techniques to change the image pixel value and position.Image compression primed by removing redundant or irrelevant information to compress images.In this paper,the neural network model of variational auto-encoder is used to compress and encrypt the image.Firstly,the best iteration times of the model is set according to the training time,reconstruction effect and loss function.Secondly,a multi-layer neural network of variational auto-encoder is used to implement the compressed image and compare with other compression methods.Finally,two standard images are used to train the variational auto-encoder generative model and extract the weights and biases,and then the data corresponding divide.The processed data load into the variational auto-encoder generative model to encrypt and decrypt images.The main work and contributions are as follows:For the problem of image encryption,a framework based on variational auto-encoder generative model image encryption is proposed and verify the model has a good compression effect.Specific steps are as follows:Firstly,the image is converted into an operational data matrix and normalized.Secondly,a variational auto-encoder network model is built and train and obtain a probability matrix.The experiment use a gradient descent algorithm to iteratively update weights and biases,and reconstruct a high-definition image,and then determine the number of iteration times.Finally,the weights and biases of two different images in the trained generation model divide the corresponding data and load it into the generative model to generate the encrypted image.The experimental results show that the image compression based on the variational auto-encoder can reduce the bandwidth of data propagation and the image encryption method based on the generative model is easy to implement and can be effectively applied to image encryption,and the decryption image of distortion rate is low.