Edge based probabilistic relaxation for sub-pixel contour extraction

Abstract

The paper describes a robust edge and contour extraction technique under two types of degradation: random noise and aliasing. The technique employs unambiguous probabilistic relaxation to distinguish features from noise and refine their spatial locations at sub-pixel accuracy. The most important component in the probabilistic relaxation is a compatibility function. The paper suggests a function with which the optimal orientation of edges can be derived analytically, thus allowing an efficient implementation of the relaxation process. A contour extraction algorithm is designed by combining the relaxation process and a perceptual organization technique. Results on both synthetic and natural images are given and show effectiveness of our approach against noise and aliasing.