Optimizing Region of Support for Boundary-Based Corner Detection: A Statistic Approach

Wen-Bing HORNG  Chun-Wen CHEN  

IEICE TRANSACTIONS on Information and Systems   Vol.E92-D    No.10    pp.2103-2111
Publication Date: 2009/10/01
Online ISSN: 1745-1361
DOI: 10.1587/transinf.E92.D.2103
Print ISSN: 0916-8532
Type of Manuscript: PAPER
Category: Pattern Recognition
corner detection,  discriminant criterion,  optimization,  region of support,  variance analysis,  

Full Text: PDF>>
Buy this Article

Boundary-based corner detection has been widely applied in spline curve fitting, automated optical inspection, image segmentation, object recognition, etc. In order to obtain good results, users usually need to adjust the length of region of support to resist zigzags due to quantization and random noise on digital boundaries. To automatically determine the length of region of support for corner detection, Teh-Chin and Guru-Dinesh presented adaptive approaches based on some local properties of boundary points. However, these local-property based approaches are sensitive to noise. In this paper, we propose a new approach to find the optimum length of region of support for corner detection based on a statistic discriminant criterion. Since our approach is based on the global perspective of all boundary points, rather than the local properties of some points, the experiments show that the determined length of region of support increases as the noise intensity strengthens. In addition, the detected corners based on the optimum length of region of support are consistent with human experts' judgment, even for noisy boundaries.