A novel watermark embedding technique based on human visual system

Abstract

Digital multimedia data are easy to be reproduced and distributed over Internet. Therefore digital-watermarking technique is needed to provide owner``s identification to the multimedia data to protect them. In general, better watermark must be perceptually invisible, undetectable without key and robust to modification of spatial/temporal data or transform. To meet these requirements, in this paper, we use the characteristics of the human visual system(HVS)where the response of visual cortex decomposes the image spectra into perceptual channels that are bands in spatial frequency. This HVS property has been employed in the MPEG-7 texture descriptor recently. First, the spatial frequency domain is divided into 30 channels which are composed with five octave-bands division in radial axis and six equal-width divisions in orientation axis. Based on the 30 channels, a watermark is inserted into the channels which have same number of bits of the watermark. In doing so, more bits of the watermark are embedded into significant portions, which locates low-middle frequency region in the frequency domain. Above watermark insertion scheme gives a robustness because the watermark of significant portion can alive against strong attack of image. In our experiments, the proposed method is robust to various attacks such as JPEG compression, filtering, cropping, rotation, scaling, multiple watermarking and so on. This results from embedding relatively more watermark in perceptually significant portion according to the information of each HVS band in the spatial frequency domain. The performance of the proposed algorithm compared with other transform domain watermarking methods has been analyzed, using various test images. As the detector response has higher correlation in the same PSNR, we have verified that the proposed method is better than other methods.