Research On Local Feature Based Reversible Data Hiding In Spatial Domain

Although information technology greatly changes the social production mode,the information security problem becomes increasingly fatal.In order to guarantee the safety of data storage and transmission,as well as reduce the exposure risk of secret data in the channel,data hiding technology is developed and has become one of the most important research topics in the field of information security.Reversible data hiding(RDH),being able to restore the cover image and secret data without distortion at the decoding end,plays an important role in telemedicine,satellite communications and other scenarios that require lossless cover image.However,how to effectively explore the redundancy in the cover image to embed more secret data in a limited number of cover images,and ensure the high-fidelity of embedded image is still a key research problem.From the aspect of image local features,this thesis proposes new methods within the framework of low-capacity or high-capacity reversible data hiding,and mainly includes three works as followings:(1)Adaptive complexity for pixel-value-ordering(PVO)based reversible data hiding.PVO algorithm is a widely used reversible data hiding framework which aims to achieve the high quality of embedded image under low capacity.However,in the existing PVO-based RDH methods,the context pixel selection is too simple or even rough,which leads to a low pixel-selection accuracy.To solve this problem,a novel complexity based on adaptive context pixel selection strategy is proposed.Different from the block-based complexity in the previous PVO-based methods,our proposed method adaptively selects context pixels from the perspective of the relative locations of both predicted pixel and prediction pixel.Consequently,instead of sharing the same block complexity by two predicted pixels in the current block,each predicted pixel can be utilized independently according to its own corresponding complexity.Moreover,different number of high correlation context pixels can be adaptively selected and the distri-bution of local pixels is measured.Our proposed adaptive complexity can combine with other PVO-based methods and the experimental results show that our proposed method achieves a significant improvement in prediction accuracy and embedding performance.(2)Multi-complexity fusion mechanism for pixel-value-ordering based reversible data hiding.In existing PVO-based RDH methods,there are two main context pixel selection strategies:block-inside and the block-outside pixel selection strategy,which show the strong complemen-tary characteristic.To make full use of this characteristic,a multi-complexity fusion mecha-nism is proposed.First,we treat the complexity problem as a binary classification problem,and propose a linear weighted fusion mechanism.To achieve the best classification performance,we take the average precision as the optimization objective.Moreover,based on our previous research results,a simpler but effective complexity,which is named as Neighboring Complex-ity,is proposed.By fusing different complexities,which represent different local features,our proposed method can effectively take advantage of each complexity,and obtain a higher clas-sification precision.Experimental results show that by applying the multi-complexity fusion mechanism to PVO-based algorithms,the embedded distortions are all reduced.(3)Reversible data hiding for high dynamic range images using two-dimensional prediction-error histogram of the second time prediction.High dynamic range(HDR)image has become a new image standard in multimedia communications because of its better lighting and detail performances.We proposed a novel RDH scheme for HDR images based on the local cor-relation among channels.Firstly,the prediction method is optimized according to the RGBE encoding format of the HDR image.While effectively exploiting the redundancy among chan-nels,the original information is protected from the impact of data embedding.Secondly,a second time prediction method is designed to make full use of local texture similarity among channels and improve the prediction accuracy.Moreover,a two-dimensional prediction-error histogram(PEH)is used in the prediction process to organize prediction-errors and maximize the embedding capacity.Finally,we proposed an adaptive embedding strategy to optimize the embedding parameters and reduce the embedding distortion.Experimental results show that our proposed method can significantly increase the embedding capacity of HDR images and obtain high-fidelity embedded images.