Research On Reversible Data Hiding Technology

Reversible data hiding technique is a process that embeds secret information into host media in a reversible, lossless way. The so-called reversible, lossless means not only the secret information can be correctly recovered, and the media can also be recovered completely, without any damage. The technique is usually used for content authentication, trademark protection, secure document, military communication etc.Two kinds of common reversible data hiding algorithm in the existing literature are: histogram-based algorithm and difference expansion-based algorithm. The histogram-based algorithm modify the histogram of host image for data embedding, in which the host image changes little and has low distortion. But, the algorithm performance completely depends on image histogram, and for the reason of gentle image histogram, the image capacity is usually low. Difference expansion-based algorithm is the main algorithm now, the core idea of this algorithm is to calculate the difference between original pixel value and its predicted value, and expand the difference two times for data embedding. In the difference expansion-based (DE) algorithm, image pixel value prediction is a crucial step which will to a large extent affect the embedding performance. Small prediction error for an image pixel often means large embedding capacity and low image distortion. But in the existing literature, people pay more attention to how to reduce the size of auxiliary map and the size of header information, ignore the importance of pixel prediction. Common prediction algorithms are just direct replacement by a neighboring pixel, mean of the neighboring pixel values, etc. Most of these methods do not take into account the different types of image content and hence leave more accurate adaptive image value prediction methods to be sought.Under this background, we propose a new adaptive reversible data hiding method through autoregression and a data embedding framework based on the prediction error histogram modification. Unlike conventional data hiding techniques, a threshold is adjusted for each image to divide all pixels into two regions:the smooth region and the texture region, then the proposed algorithm exploits the correlation inherent among the neighboring pixels in an image, optimally estimates the coefficients of the autoregression model for image pixel value prediction by least-squares minimization. For the reason that different images have different thresholds and different regions in an image have different prediction coefficients, the algorithm completely breaks the limit of traditional algorithms and fully takes different image content into account, which leads to higher embedding capacity and better image quality. Experimental results show that the adaptive prediction algorithm offer valuable advantages over state-of-the-art methods in general. A general data embedding framework in the prediction error domain through histogram modification proposed by us can be viewed as a natural extension from the past work using histograms. Analysis shows that some data embedding methods, including the DE approach, which is one of the best in performance, are special cases in this framework.