Study On High-capacity, Low-distortion Image Reversible Data Hiding

Data hiding is an emerging multi-disciplinary field technique in information security. Data hiding technique hides secret information into public digital media through a certain method under the premise of not affecting the normal use of the original host media. Because this technique itself has many advantages beyond the traditional cryptography, data hiding technique has been widely used in digital media copyright protection, covert communication and other fields. As an important branch of data hiding,reversible data hiding technique can extract the embedded secret message correctly and recover the original host media without any distortion at the same time. The technique solves the problem of permanent damage to the host media due to the data embedding in the conventional data hiding technique, and has become a new hot research direction of data hiding field. For some special application areas, such as medical, legal, military applications, even subtle distortion of image content may cause erroneous judgement to users, so this host media lossless-restoration characteristic of reversible data hiding technique becomes particularly important.Image is one of the most commonly used information carrier of human social activities, taking this as the background, this paper mainly focuses on high capacity, low distortion, digital image based reversible data hiding algorithm. The main contributions include:1. A histogram expansion based universal reversible data hiding framework is proposed. On the basis of deeply exploring the evolution process and core idea of histogram shifting based, prediction-error histogram shifting based and prediction-error expansion based reversible data hiding methods. This framework clarifies the similarity relation between prediction-error expansion based and histogram shifting based reversible data hiding methods, summaries the basic idea of histogram expansion based reversible data hiding method: shift certain pixels in any histogram domain to create vacant space whereas some others are expanded to embed secret message. The framework includes more than one existing methods as special cases, has great flexibility and versatility as well as important guiding significance for future design of histogram expansion based reversible data hiding algorithm.2. An adaptive reversible data hiding method through autoregression is proposed.The method utilizes the variance of supporting region to measure pixel neighborhood flatness and classifies pixels by optimal threshold parameter, then uses more accurate and reasonable autoregression model to predict the pixel values in the smooth region and texture region respectively. Compared with the traditional causal window based pixel value prediction methods, this method improves the prediction adaptability and achieves higher prediction accuracy. On this basis, the capacity-distortion performance of conventional reversible data hiding algorithm is improved.3. A novel multi-resolution reversible data hiding framework is proposed. The framework uses an image pyramid based multi-resolution embedding rule to converts the image pixel value prediction problem to the classical image interpolation problem, thus the state-of-the-art image interpolation(zooming) methods can be utilized to the prediction-error expansion based reversible data hiding algorithm naturally to replace current wide used local small-scale pixel value prediction methods. By using the advantage of preserving image textual and structural details of image interpolation algorithm, combining with pixel sorting/selection strategy, this framework improves the capacity-distortion performance of conventional reversible data hiding algorithm.4. A framework for image interpolation using weighted surface approximation is proposed. The framework converts the weighted mean square error of low-order polynomial approximation to a continuously distributed probability of a pixel belonging to a local smooth region or a textural one, thus essentially making a soft pixel classification.Meanwhile, the prediction under smooth region assumption is obtained by low-order polynomial weighted approximation, combining with the existing method for prediction under texture region assumption, superior image interpolation result is produced by comparing with current mainstream methods. In addition, this framework can be directly applied to the above proposed multi-resolution reversible data hiding framework to improve the pixel value prediction accuracy and further optimize the final performance of reversible data hiding algorithm.5. A low-distortion reversible data hiding algorithm using DCT coefficients energy analysis based prediction method and optimized embedding strategy is proposed.The algorithm utilizes the characteristic of energy distribution of natural image pixel block DCT coefficients to predict pixel value, thus improves the prediction accuracy of conventional method. Moreover, the algorithm proposes Pure load-Image distortion function, which can be used to find optimized pixel shifting direction, secret message embedding location and pixel sorting/selection threshold in every embedding in the image blocking multi-embedding process to better handle the contradiction between secret image quality and data hiding capacity. Experimental result shows that for low data hiding application, comparing with current mainstream methods, the proposed algorithm always produces high quality secret image in the same secret message embedding capacity.