Research On High Performance Reversible Information Hiding Method With Multiple Feature Constraints

In the age of digital communications,data security and communication security are top priorities,with new technologies emerging daily and the proliferation of the Internet making information sharing easy and effortless.However,data protection and security is a constant challenge because information is shared over insecure networks.Traditional data hiding techniques usually cause irreversible distortion of multimedia data and therefore cannot be used for applications that do not require permanent distortion of the original image,such as medical and military image processing.In recent years,reversible information hiding has been extensively studied to cope with these highly sensitive scenarios.In this thesis,we explore the pixel prediction process of reversible information hiding algorithms in depth and construct a multi-feature constrained prediction mechanism using multi-layer perceptual features of images,which can help to provide higher performance of reversible information hiding algorithms.The research in this thesis includes two main aspects.First,this thesis proposes a multi-feature constrained prediction mechanism oriented to pixel value ranking algorithm,which can make full use of the complexity of pixel texture and the method of effectively fusing different prediction patterns,introducing a prediction context constraint on pixels using different prediction patterns,containing 16 prediction patterns,which helps to select the set of predictions with the best embedding performance from the surrounding pixels,combined with local complexity constraint,in addition to the restriction on the order of embedded pixels,the data embedding is also achieved using multi-prediction error histogram shifting based on the local complexity histogram.Compared with the traditional algorithms,the effectiveness of the proposed algorithm is verified on the USC-SIPI image dataset and Kodak image dataset,which are widely used nowadays.Secondly,in order to improve the performance of reversible hiding algorithm for color images more efficiently,this thesis proposes a reversible information hiding algorithm for color images based on channel correlation,which designs a fully closed prediction model based on texture and channel correlation and an embedding selection model based on edge detection as a way to construct a framework for pixel prediction with multiple feature constraints in color images.Specifically,this thesis first introduces the correlation between pixel texture and channel among multiple channels in color images,and each pixel is selected with a different type of prediction set,and then an embedding selection model that can select whether the pixel can be embedded or not is designed based on the local complexity combined with edge detection.The experimental results show that the algorithm has a great superiority in RDH of color images compared with existing similar algorithms,and it gets a great improvement in PSNR and embedding capacity.