Regression of SVM based robust watermarking algorithm

This paper presents a robust watermarking algorithm based on regression of SVM. The host image is divided into two parts firstly. The first one is to train SVM using features including four relationships between the neighbor pixels, and the other part is to embed watermarks. The center pixels of image blocks are compared with the predicted pixels of trained SVM, and then modified to embed watermarks. Because of the introduction of Pseudorandom Circular Chain (PCC) scheme, the training pixels and the test pixels can spread all over the host image, which increases accurate rate of the watermark-extraction. Otherwise, for the security consideration, logistic mapping is used to permute the watermark image. Experiment results demonstrate that this algorithm can get good perceptual invisibility, and robustness against many attacks.