Fuzzy Linking Models for Pyramidal Edge Detection

This paper presents a novel approach to multiresolution edge detection, which combines the grayscale morphological filtering, pyramid data structure and fuzzy technique. It mainly addresses the linking of edge nodes at adjacent levels in image pyramid. In previous pyramidal approaches, linking is based on linear relationship and intensity proximity only. The approach proposed here contains multiple linking mechanisms and introduces fuzzy technique. It considers the parent-child linking relationship of edge nodes between the two adjacent levels as fuzzy model, which is trained offline using real image data. Through this fuzzy linking model, the coarse, low-resolution edge map is propagated and refined to the fine, high-resolution edge map in the pyramid. The validation experiment is carried out on one synthetic image and two real images, and the results show that our approach has better performance on the localization and detection of continuous large-scale object boundaries than Canny’s edge detector and other previous multiresolution approaches. In addition, the proposed approach has high computational efficiency.