Image watermarking algorithm based on grey relational analysis and singular value decomposition in wavelet domain

Abstract

An image watermarking algorithm based on grey relational analysis and singular value decomposition in wavelet domain is proposed. Firstly, the host image is processed with one-level of discrete wavelet transform. The low frequency coefficients LL1 can be obtained from mentioned operation, and LL1 is divided into non-overlapping blocks whose size is same as watermarking. Secondly, through the gained coefficients of each block and the given random sequence, grey relational degrees which are preserved as training sample are acquired for each block. The largest singular value which can be found from singular value decomposition for each block is preserved as training target. Thus total training samples and corresponding training targets are obtained. Then, The LS_SVR model can be obtained through the training study. Next, through feeding the trained LS-SVR with the training samples to estimate the largest singular values, watermarking bits are embedded for adjusting the largest singular values. Finally, the watermarking is extracted by the reversing steps, and the extraction algorithm belongs to non-blind watermarking because the original host image is necessary. Experimental results show that the proposed scheme not only possesses good imperceptibility, but also has fine robustness against common signal processing.