Research On Reversible Watermarking Techniques

In the past decades, digital watermarking has been widely used in areas such as ownership protection, content authentication, distribution tracking, and broadcast monitoring. In most cases, the cover media will experience some permanent distortion due to watermarking and cannot be inverted back to the original media. However, in some applications, especially in the medical, military and legal domain, even the imperceptible distortion introduced in the watermarking process is unacceptable. Under these circumstances, reversible watermarking is desired, which not only extracts the watermark, but also perfectly reconstructs the original host signal from the watermarked work. In this thesis we investigate several problems on reversible watermarking.Some creative productions are given as follows:(1) Based on difference expansion and difference shifting, a novel reversible watermarking algorithm was proposed. A sequence of pixels was formed by scanning the image in a given order, and the difference of every adjacent pixel-pair fell into three parts: one was shifted right if the difference was bigger than K-1; another was shifted left if the difference was smaller than -K; and the third part was used to embed the data (K was a positive integer chose by the user). At the same time the distortion caused by expansion was uniformly separated into the two pixels, which kept the high quality of the image. Experimental results demonstrate that the proposed scheme can embed a larger payload while keeping the same PSNR (peak signal-to-noise ratio).(2) A novel reversible watermarking scheme was proposed, which was robust against high-quality JPEG compression. By computing the difference between the means of two adjacent image blocks, employing it to embed the watermark, and using block histogram shifting to avoid overflow/underflow, this technique has received both losslessness and robustness. If the watermarked image does not change at all, the watermark can be extracted out correctly and the original image can be recovered losslessly. If the watermarked image is compressed to some extent, the watermark can still be correctly extracted for semi-fragile authentication. The scheme has been successfully applied to three kinds of medical images and commonly used images. Experimental results show that the high visual quality of images, the pure payload, and the robustness of the proposed scheme against compression are acceptable for many applications, including semi-fragile image authentication.(3) Based on difference a novel reversible watermarking algorithm for digital vector map was proposed. The watermark was embedded by modifying the vertices coordinates. Instead of difference expansion, the difference of every adjacent coordinates fell into two parts: one was shifted further away from 0; another was used to expand and embed the watermark (K was a positive integer chose by the user). At the same time, the distortion caused by expansion and shifting was uniformly separated into the two pixels, which kept the lower distortion. Experimental results demonstrate that the proposed scheme can embed a larger payload while keeping the lower root mean square error (RMSE).(4) We propose a novel reversible watermarking scheme for SVG files. Reversible watermarking is suitable for hiding data in 2D vector graphics, especially in Scalable Vector Graphics (SVG) medical images, because it can remove the distortions induced by watermark embedding after extracting the hidden bits. Since the 'path data' in SVG files includes a sequence of line segments and Bézier segments, we embed the watermark into the vertex coordinates of 'path data' by modifying the differences between the adjacent vertex coordinates. The scheme can keep the Bézier segments c~0 continuous by unchanging their start points and end points, and need record neither the location map nor the original least significant bit (LSB) of difference. Our experiment results show that the proposed scheme achieves good performance of the imperceptibility and high-capacity data embedding.(5) A new algorithm based on difference is proposed by using the traversed neighbors of an element in the 3D mesh models. To increase the robustness of reversible watermark, we just use difference shifting to embed the watermark. Experimental results show that the watermark can be extracted out correctly if the watermarked model does not change at all, and the original model can be recovered losslessly, the watermark can still be correctly extracted even after the watermarked model was attacked by adding noise, filtering and slightly rotating.