R. H. Yan, N. Tokuda and J. Miyamichi
Utsunomiya University, 350 Mine Machi, Utsunomiya, Tochigi 321, Japan
(Received April 10, 1996; Accepted July 29, 1996)
Keywords: Sharp Recognition, Corner Detection, Sharp Smoothing
Abstract. This paper presents a new, robust corner detection algorithm applicable to digital curves of noise-polluted boundaries of an object. The method extends Pikaz and Dinstein's smoothing algorithm based on feature point detection directly to corner detection with an areal comparison added in the stop condition for an improved feature extraction accuracy. We have also proposed a new corner sharpness function whereby those corners exceeding a certain threshold value are recognized as a corner even if they satisfy the common definition of a local maxima in curvature. We have presented experimental verifications that our algorithm is robust to noises and that it produces corners of objects in agreement with our more intuitive human object recognition processes.