New Edge Detection Approach by Using Linear Filtering

This paper proposed a new method for edge detection based on the diffused images that are obtained by diffusing the source image with different parameters of filter, respectively. The edge map is determined by checking the varying of the pixel value on the diffused images, and multiscale edge detection can be achieved by using different filter parameters. Unlike previous approaches, the new method is robust to the varying of some parameters of images such as contrast, brightness, etc. Also it is computationally very efficient; in particular when the linear diffusion methods like four-points averaging algorithm is applied. Furthermore because the edges or contours of the image is located in the edge map based on the principle of level set-based segmentation algorithms, therefore the detected result is the same with that in the segmentation algorithms. The key to this algorithm is the determination of the filter parameters, and it can be tried and determined in advanced for a specified application. The proposed algorithm has been validated with some numerical results.