A Framework for Bean-Shape Contour Extraction

Contour extraction and object detection is one of fundamental problems in computer vision. Contour extraction can be guided by either global or local constraints. In this paper, we propose a local constraint based framework for bean-shape contour extraction. We propose a criterion to construct primal sketches based on connected components of Canny edge points in a channel-scale space. When a targeting object is surrounded by a complex background, a sketch token may be deficient (not closed), and it may also contain some faulty part (not on the boundary of a targeting object). We propose algorithms to detect and restore deficiencies and faults of primal sketch tokens. We present two case studies for the proposed framework: i) embryo localization, and ii) face localization. The case studies demonstrate the potential of the proposed framework.