Research On Image Reversible Data Hiding Method Driven By Convolution Neural Network

In recent years,with the development of computer science and network communication technology,the communication based images has become more and more important.Consequently,ensuring the security and integrity of image information in the process of sharing and transmission has become a key issue in the field of cyberspace security in recent years,which has extremely important value and significance.Image reversible data hiding is a technique for embedding data into cover image.The original cover image can be recovered lossless while extracting the embedded data,which ensures the invisibility,unpredictability,robustness and security in data hiding,and the integrity of the original cover image.It has high confidentiality and fidelity,and is highly suitable for military,medical,judicial,copyright protection and other fields.The prediction error expansion is important to the current reversible image data hiding methods.Because it can ensure the quality of the embedded image and has high embedding capacity,it is widely studied.With the breakthrough of deep learning in images,pixel prediction has been developed rapidly,which has important implications for building a high-performance convolutional neural network predictor,improving the prediction accuracy in the process of reversible data hiding of images,and increasing the space available for reversible information hiding of encrypted images.Therefore,this dissertation applies the pixel prediction technology driven by convolution neural network to image reversible data hiding method,and carries out adaptive improvement and multi-scene verification.The main research work are explained as follows:1.A Convolutional Neural Network driven Optimal Prediction for Image Reversible Data Hiding method is proposed.The current typical pixel prediction methods are highly dependent on the neighboring pixels of the predicted pixel,which can not make full use of the correlation between pixels,affecting the performance of the cover image and the capacity of embedded data.In order to make up for the above defects,this dissertation designs a predictor based on convolution neural network to improve the prediction accuracy of image pixels.At the same time,in order to make up for the shortcomings of the convolutional neural network predictor,this dissertation designs an optimal prediction performance algorithm,so that the convolutional neural network predictor can be applied to image reversible information hiding adaptively.The experimental results show that the method proposed in this dissertation can obtain the sharpest prediction error histogram,and has better performance.2.A reversible data hiding method in encrypted images driven by convolution neural network is proposed.The reversible data hiding in the encrypted images is very difficult to make room for the data to be embedded,which greatly affects the embedding capacity and the embedding effect.To solve this problem,this dissertation proposed a reversible data hiding method in encrypted images driven by convolution neural network,which designed a convolutional neural network predictor to make full use of the correlation between the pixels of the original cover image,improve the effective embedding capacity.At the same time,the structure and parameters of convolutional neural network are used as encryption keys to enhance the security.The experimental results show that the method proposed in this dissertation can achieve the best performance of the encrypted image quality under general conditions,and the performance advantage is more obvious with the increase of the embedding capacity.