Automatic estimation of dynamic filter scales for multiscale gradient multiplication edge detector

Edge detection is one of the fundamental tasks in image processing and computer vision. Multiscale gradient multiplication (MGM) edge detector can improve the edge localization accuracy. However, the existing MGM edge detector only adopts two fixed filter scales, which can not self-adaptively deal with the images with different signal-to-noise ratio. This paper proposed an automatic estimation theory of dynamic filter scales to improve the traditional MGM edge detector. The automatic estimation theory of dynamic filter scales is based on the different feature of the first derivative of Gaussian responses to the edges and the noises. The experimental results on both noisy synthetic and natural images show that the proposed scheme is robust and has superior edge detection performance.