A contour extraction method of coke optical texture based on edge confidence map

Contour information is an important feature in recognition of coke microscopic optical texture. In view of the characteristic of coke microscopic, a contour extraction method, which is based on joint edge confidence map and dual threshold, is proposed. At first, an edge confidence map with gradient magnitude and orientation is built, then non-maximum suppression is employed to obtain local external points; by setting dual thresholds and judge location of points, appropriate edge points are obtained; by tracking these edge points, contour is formed. In this method, parameters can be auto-adjusted. Experimental results show, with the proposed method, the weak contour can be detected, continuity and fine details of contours of different optical texture components can be preserved, which is an important basis of subsequent contour feature extraction and corresponding optical texture component recognition.