<title>Digital image watermarking by salient point modification: practical results</title>

Abstract

A new watermarking method is presented for still images and video streams. The method is different from nearly all known methods for image watermarking, which are based on adding pseudo-random noise to luminance or color components of the pixels. The new method is based on biasing the geometric locations of salient points in an image. The watermark is formed by a pre-defined dense pixel pattern, such as a collection of lines. So-called 'salient points' in the image are then modified, e.g. by warping or by changing the local luminance pattern around a salient point, such that after watermarking a majority of the new salient points lies on the watermark pattern. This paper describes the details of the new watermarking method and discusses the results of a series of tests performed on watermarked images. The feasibility and robustness of the method are shown.