A two-step approach to detect contours formed by sharp intensity changes

This paper suggests a two-step model to ease a curve detection problem. Instead of linking edge detection results into segment-like structures, we embed curve detection into an edge detection process. The method consists of two major steps, namely, (1) formation of curve-associated region for each curve and (2) curve extraction from each curve-associated region. A curve-associated region is an extended area surrounding a curve where associated intensity variations of the curve are preserved. Gradient orientation is the main attribute used to group pixels into curve-associated regions. The intensity structure of each curve-associated region is then analyzed to locate the curve. This method appears to be a viable approach to complement edge detection techniques because of the utilization of concepts, such as, (1) use of gradient orientation, (2) integration of local information over space, and (3) analysis of global intensity structure of each curve