Subsampling-Based Wavelet Watermarking Algorithm Using Support Vector Regression

Abstract

A subsampling-based wavelet watermarking algorithm by using support vector regression (SVR) in the wavelet domain is presented in this paper. Four coefficient sets are obtained via DWT for four subimages gained by subsampling an original image. Because of the neighborhood correlation of image pixels, the coefficient sets are approximately equal. Due to the good learning and generalization capability in the processing of small-sample learning problems, SVR is applied to model the relationship between the coefficient on the random selected coefficient set and the coefficients on the corresponding position of others. Then, the watermark is embedded into part of the low frequency coefficients or extracted by adjusting or comparing the relationship between the embedding coefficient and the output of the trained SVR. Experimental results show our technique has excellent performance against several common attacks.