Research On Reversible Digital Watermarking Based On Prediction

The research of reversible watermarking algorithm has been widely studied as a new direction in the field of information security recent years. The traditional digital watermark embedding process will usually bring embedding distortion that cannot be erased to the original carrier. Although the distortion is usually very small, the degree of distortion is still not acceptable for some image applications like medical or military field. In order to solve the problem, some scholars put forward the reversible digital watermarking technology. Its main thought is to embed data or add watermark in the original information carrier, and recover the undistorted original carrier information. Its main advantage is that the user can extract the embedded useful information from the embedded data or adding watermark carrier signal at the decoding side, and the user can also resume the original carrier signal without distortion.Based on a lot of analysis of existing research results, the main innovations of this paper are as follows:The first innovation:Multi-dimensional prediction-error expansion(PEE) algorithm. The traditional PEE is a widely used technology, the deeper research is just pairwise PEE, fail to take full analysis and advantage of the correlations among prediction-errors. In this paper, the PEE is expanded to a prediction-errors group, by abandoning the large-embedding-distortion-mapping to reduce the total distortion. Then the experiment illustrates the multi-dimensional prediction-error expansion is better than pairwise PEE in performance, and more suitable for simple and smooth texture images. This algorithm is simple and superior.The second innovation: reversible watermarking by combing local and global research. Finding an appropriate predictor is the first and also the most important step for the histogram shifting based reversible data hiding. Unfortunately, most predictors reported in the literature cannot be directly introduced to the embedding process. To this end, this paper proposes a hybrid predictor for histogram shifting, which not only uses the local information near a pixel, but also utilizes the global information of the whole image. In addition, we design an estimation function which enables the use of sorting.As a result, the embedding performance is significantly improved. The superiority of the proposed data hiding method is finally experimentally verified by comparing with four state-of-the-art methods.