Corner detection by local histograms of contourchain code

A new method for determining the shape of an object by obtaining its dominant points or border corners is presented. The algorithm uses the contour chain code as descriptor and calculates the curvature of each border point, comparing two histograms of the chain code. The method is not only fast, with the detected corners stable to rotations and scaling, but also the curvature function presents a high signal-noise relationship, and thus offers excellent performance for real contours.