Abstract: Aiming at the shortcomings of the current digital image watermarking algorithms, such as poor information hiding performance, low security and easy attack, we proposed digital image watermarking algorithms based on multi\|scale transform and support vector regression. Firstly, wavelet transform was used to decompose the original digital image into many different subbands. Secondly, subband coefficients were trained by support vector regression, and the relation between subbands was estimated. Finally, watermark was embedded into image based on the results of training support vector regression machine, and its performance was tested through the specific digital image watermarking simulation experiment. The experimental results show that the proposed algorithm has good information concealment and good imperceptibility, and it can resist all kinds of attacks and improve the security of digital images.

Key words: wavelet transform, support vector regression, information concealment, watermarking algorithm