Video Compression And Encryption Algorithm Based On Hybrid Neural Network

With fast development of multimedia information technology and computer network, multimedia data which can be expediently and fast transmitted in network gradually become the most important source of information. Compared with other kinds of multimedia data, video which has many advantages such as visual and intuitive performances, and also could cover more information, has turned into a primary communication method. However, considering the easy understandability of video data, anyone could get the information according to the data, so how to protect the security of video information is becoming the focus on the study of scholars all around the world.To further improve the performance of video encryption by integrating the video compression and encryption, this paper proposes the feasibility of video compression and encryption integration based on hybrid neural network, the main researching job of this dissertation is as follows:Based on the coding structure framework of video compression standard H.264, considering the characteristics of RBF neural network such as non-linear and excellent ability of digest, a novel method is proposed to achieve the video compression and encryption integration by substituted the intra-prediction and inter-prediction modules with RBF neural network. Because of a large quantity of weight data of the neural network and good performance of SOM neural network in vector quantization, this scheme chooses SOM neural network to compress the network weights of the sub-RBF neural network for decryption and decompression, the codebook generated by the vector quantization is transferring as the new key. To satisfy the requirement of real-time video transferring, a hybrid learning algorithm based on Nearest Neighbor Clustering Centers-selected Algorithm and Gradient Descent Training Algorithm is used to train a RBF neural network, at the same time, an improved learning method is applied to train the SOM neural network, these new methods could reduce the training time. Experimental results show that this algorithm possesses the features of high security and no impact on compression ratio. Moreover, it could speed up the process of video compression and encryption.