Enhanced Edge Detection Technique for High-Resolution Satellite Imagery Using a Curvelet-Canny Approach

Abstract

In this research paper, a new approach using the capability of the second generation curvelet transform together with the traditional canny operator for edge detection of highresolution satellite imagery; the combined technique will be applied on WorldView-2 imagery. First, the curvelet coefficients will be generated in multi-scales and multi-directions using a forward discrete curvelet algorithm. Then, these coefficients will be sorted in each scale to generate the edge map using the larger coefficients for the coarser scales. Second, this edge map will be the input to the second stage where the three main steps of the traditional canny operator, gradient calculation, non-maximal suppression and hysteresis, will be applied. The first step results in removing noise, fine edges, from the image aiding the second step for better connecting the strong edges without the effect of weak edges coming from the noise. The percentage of the utilised coefficients in the curvelet transforms step together with the weight for each scale are the tuning parameters the user has to adjust to get the desired level of edges detected. The results from the proposed approach were compared to the traditional canny edge detection algorithm. The results showed very good potentials for detecting elongated edges and also for generating more closed objects, which make this method a good alternative for the segmentation step for any further object-based classification algorithm.