A region-based edge detection technique for noisy images

Edge detection is a basic and important issue in computer vision and image processing. The traditional methods of image edge detection are sensitive to noise and often cannot locate the edge exactly. We propose a new method of edge detection which can eliminate the fake edges caused by the noise disturbance, and ensure the veracity of edge orientation to have a proper result for noisy images. A problem introduced as "cross effect" occurs in the derivative of Gaussian filter method, which will be eliminated in this paper. As will be shown, our method is robust in confrontation with noise.