Blind watermarking via NMF with sparseness constraint

NMF is a linear representation that could lead to sparse coding of natural images, which coincides with human being´s vision system. In this paper, the blind watermark algorithm via NMF, for the first time in the field of watermark, is proposed. Spread spectrum embedding scheme is adopted in order to make a controllable tradeoff between robustness and perceptual imperceptivity. Hoyer´s NMF (2004) approach is adopted to evaluate watermark performance under different sparseness constraints, of coefficients and image basis respectively. Investigations are carried out through experiments, and the results show that suitable sparseness constraint is essential for good watermarking, and parts-based basis are more robust than orientation, band-pass basis. Watermark extractions under various attacks are illustrated, show algorithms´ effectiveness and robustness.