Abstract: An image watermarking scheme was developed based on SVM (support vector machine) and SVD (singular value decomposition),which is used to further enhance the performance against attacks.Firstly,the host image is decomposed by the DWT transform,and its low frequency wavelet band is split into non-overlapping blocks.Then,the partial correlation model of the block is established by using the support vector machine.A feature sequence is derived through judging the numerical relationship between the prediction results and the corresponding position of the low frequency coefficient values,and the feature watermark sequence is derived through exclusive-or the feature sequence and the watermark.Moreover,the feature watermark sequence is embedded into corresponding block’s biggest singular va-lue from the original image’s low frequency wavelet band based on the principle of odd-even quantization.Finally,a watermarked image is obtained after SVD synthesis and IDWT.Experimental results show that the scheme is not only invisible but also has strong ability to resist attacks.

Key words: Digital watermarking,Support vector machine,Singular value decomposition,Odd-even quantization