Research On Reversible Digital Watermarking Of Color Image

As an important technology to protect the high fidelity and confidential images, the reversible watermarking can recover the original carrier image as well as extracting the watermark in the decoder, such that this technique will be widely applied in military, medicine and other fields. The traditional reversible watermarking methods are mainly designed for protecting the gray images. However, it has some difficulties when we extend such methods to color images which are generally adopted in real life. To this end, the main research direction of this paper is to design the more efficient watermarking algorithms for color image.This paper presents two improved algorithms of reversible watermarking of color image based on the research of the existing reversible watermarking of gray and color image:(1) The reversible watermarking algorithm of color image based on three-dimensional prediction-error histogram is proposed. Firstly, this algorithm divides the color image into three gray images, and the rhombus prediction is adopted in each gray image to obtain the prediction errors. Then, three prediction errors of every pixel in the three gray images are gathered to compose the group-prediction errors. Finally, in order to reduce the image distortion, the mapping relationship of the group-prediction errors is modified by rejecting the mapping with large embedding distortion and rebuilding a new mapping relationship. Experimental results indicate that the proposed method achieves significant improvement over some state-of-art methods.(2) The reversible watermarking algorithm of color image based on nine-rectangle-grid prediction model is proposed. On the basis of traditional prediction error expansion algorithms, the proposed algorithm modifies the prediction model with the nine-rectanglegrid prediction model composed of three color channels. According to the image texture complexity, the mixing prediction model of nine-rectangle-grid and rhombus is applied: using the nine-rectangle-grid prediction model in the area with complex texture and using the rhombus prediction model in the smooth area. As a result, the prediction accuracy of pixels is increased and the image distortion caused by embedding the watermarking is reduced.Both of the proposed algorithms can achieve an improved performance because of using the correlation among three color channels of color image. The experimental results of standard test images prove that the proposed algorithms produce less distortion than some existing embedding schemes of color image for the same embedding capacity.