A new texture-insensitive edge detection method

In this paper, a novel edge detection method is proposed which can be used to find main edges of objects meanwhile filter edges within texture regions. We first introduce three kinds of edge models and three kinds of nonedge models to represent all the possible condition in the edge detection process. Then, pixel's window intensity regularity (WIR) is defined which denotes the texture similarity degree around the pixel. And then based on pixel's WIR a new computation method for image gradient is brought forward, according to which, gradient value within texture regions will be almost eliminated while the reasonable gradient will not be weakened obviously. Finally, Canny-like operator is employed to detect and locate edges. Experimental results demonstrate encouraging performance.