Thin Edge Detector

Most of edge detectors are based on detecting points in the image with a high gradient value. However, in general, they omit the idea that the edge must be a connected line of pixels. Instead of that, these techniques normally use a global threshold value, to control the quantity of edge points detected. Due to this, many other edge points are detected, but some of them do not really belong to an edge. In this paper we propose a new edge detector, that detects entire pieces of 1-pixel thick lines. This algorithm is more robust concerning the choice of the threshold value, because it works with the average gradient value of each piece of line. Two previous filters are needed: a contrast process, to enhance the edges, and a morphological denoising to remove noise but without blurring the edges in the image. The proposed algorithm starts from pixels with a high gradient value, and follows neighbor pixels connecting new points.