Reversible Watermarking Scheme Using The Adaptively Optimized Prediction

Reversible watermarking is an important branch of reversible data hiding.By using this technology,the watermark can be hidden in a cover image.In the extraction side,not only the watermark but also the original carrier image can be recovered without any distortion.Reversible watermarking aims to embed more messages in a cover image with less distortion.The more watermarking information is embedded,however,the greater distortion would be.Therefore it is necessary to find a good compromise between embedding capacity and embedding distortion.To this end,many researchers have proposed a lot of schemes.Reversible watermarking based on prediction error expansion and histogram shifting(PEE-HS)is one of research directions that has been widely explored.For this kind of schemes,the performance mainly depends on the predictor.The more accurate the prediction is,the higher the peak at the zero bin of prediction error histogram would be,and the better the embedding performance could be.This thesis mainly focus on improving the performance of reversible watermarking by enhancing the local predictor(LP),which is the-state-of-theart predictor.As a natural image contains edges and texture and pixel values along the edge and texture are close to each other,it would be helpful to improve the predictor if the prediction is conducted along the direction of edge and texture.In light of this,the thesis proposes an least absolute shrinkage and selection operator(LASSO)-based reversible image watermarking.Specifically,according to characteristics of edges and textures in the natural image,this thesis first characterizes the prediction of image pixels as an LASSO problem,and then solves the LASSO problem to get prediction coefficients.This in turn yields both a prediction value and a prediction error.Finally,the watermark is embedded using the PEEHS.Experimental results show that the performance of the proposed scheme is better than the state-of-the-art reversible watermarking approach using the LP in case of small embedding capacity,while in the situation of large embedding capacity it is comparable to the LP-based approach.Since the LP is the-state-of-the-art predictor,the experimental results show the feasibility and effectiveness of the proposed predictor and reversible watermarking scheme.In addition,by taking into account the fact characteristics of edges and textures in natural images,this paper further designs a local predictor exploiting the content-adaptive block size,achieving higher prediction accuracy.Specifically,in constructing the prediction context,the original to-be-predicted pixel is replaced by its rhombus-averaged version,which can achieve synchronization between embedded end and extraction side.Later,a number of candidate block sizes are then set and the same number of candidate prediction errors between the rhombus-averaged and prediction values are generated.Subsequently,the same number of candidate prediction errors is obtained by subtracting the rhombus-averaged value of the pixel from the candidate prediction value,which is because subtracting the original pixel from the candidate prediction value would cause synchronization problem between the embedder and decoder.As prediction errors are actually calculated with respect to the average value instead of the original one,the prediction value corresponding to the t-th(t?1)smallest candidate prediction error is chosen as the best prediction,aiming to minimize the prediction error between the predicted value and the original one.Parameter t is obtained by experimental simulation and remains unchanged for all pixels in the image.Thereafter,the best prediction error of each pixel is used to construct a prediction error histogram,and the PEE-HS technology is used to embed the watermark.Experimental simulations implemented with the Matlab show that for any embedding rate,the proposed scheme outperforms the state-of-the-art based on the local predictor with fixed block size(LP-FBS).Compared with other schemes employing the median edge detector(MED)and the gradient-adjusted predictor(GAP),the performance of the proposed scheme is improved significantly.Also,the proposed scheme is comparable to the reversible watermarking method using the sorting and prediction.