A Novel Multiscale Edge Detection Approach Based on Nonsubsampled Contourlet Transform and Edge Tracking

Abstract

Edge detection is a fundamental task in many computer vision applications. In this paper, we propose a novel multiscale edge detection approach based on the nonsubsampled contourlet transform (NSCT): a fully shift-invariant, multiscale, and multidirection transform. Indeed, unlike traditional wavelets, contourlets have the ability to fully capture directional and other geometrical features for images with edges. Firstly, compute the NSCT of the input image. Secondly, theK-means clustering algorithm is applied to each level of the NSCT for distinguishing noises from edges. Thirdly, we select the edge point candidates of the input image by identifying the NSCT modulus maximum at each scale. Finally, the edge tracking algorithm from coarser to finer is proposed to improve robustness against spurious responses and accuracy in the location of the edges. Experimental results show that the proposed method achieves better edge detection performance compared with the typical methods. Furthermore, the proposed method also works well for noisy images.