A Novel Watermarking Technology Based on Posterior Probability SVM and Improved GA

Shiqin Liu, Minjun Zhao, Jixin Ma, Jiangyuan Yao, Yucong Duan, Xiaoyi Zhou

Abstract: The widespread distribution of multimedia data cause copyright problems for digital content. This study makes use of digital image watermarking technology to protect copyright information, and proposes a scheme utilizes the support vector machine (SVM) based on posterior probability and the optimized genetic algorithm (GA). Firstly, each training image is divided into sub-blocks of 8 * 8 pixels, and they are trained and classified by the SVM to obtain the adaptive embedding strength. Secondly, after the operation of reproduction, crossover, mutation, the genetic algorithm generates new individuals in the search space by selection and recombination operators to optimize the objective function, and find out the best embedding position of the watermark. The 8 * 8 pixel sub-blocks were transformed by DCT when embedding. Finally, the watermark is extracted according to the embedding rules. Compared with the experimental results of other algorithms, the proposed scheme has better resistance against some common attacks, such as Histogram Equalization, Guassian Noise (0.04), Guassian Noise (0.05), JPEG (QF = 50), Salt-pepper Noise (0.01).