SVR-parameters selection for image watermarking

Abstract

An image digital watermarking technique using support vector regression (SVR) is proposed and researched in this paper. Firstly, the method of embedding and extracting watermarking from digital image is given. Then, the influence of SVR-learning parameters on the watermarking performance is analyzed, and the ideal value range of SVR-learning parameters for different images is given respectively. Finally, the results are validated with other images. Experimental results show that this technique can obtain good watermarking performance as well as good learning performance when RBF kernel is adopted with its width a from 8 to 10, balanceable parameter C from 0.8 to 1, insensitive parameter s from 0.008 to 0.01 respectively