Studies on Detection of Position of Watermark

Abstract

An image processing method to detect the position of a watermark is proposed. This is one of methods utilizing pattern matching techniques. One of its characteristics is to make a watermark in an input image included within limits of a reference image, whenever the reference image as a template shifts. Then, the position of a watermark in an input image can be detected by finding points where the convolution between an input image and a reference image has a maximum value, if all pixel values of the part except for a watermark in an input image are zero. The other characteristic is that a multi-level image is made by adding bi-level images after erosion.
After each of two bi-level images drawn by a computer, i. e. an input image and a reference image, was turned into a multi-level image, the convolution between these two multi-level images is implemented whenever the reference image shifts. Results of a computer simulation make it clear that the new method utilizing the convolution between two multi-level images makes the distribution of points where the convolution has a maximum value narrower than the conventional method utilizing the convolution between two bi-level images. Experimental results with an input image captured by an image sensor camera agree with those of a computer simulation. Therefore, it is found that the new method improves the precision of the position detection of a watermark compared to conventional methods.