Contour Matching Based on Local Curvature Scale

Image matching based on contour is an important issue in computer vision, navigation and pattern recognition. The image matching methods like curvature-based methods and corner-based methods have poor robustness to the contour´s noise and distortion, and some matching methods are applied only to closed contours. A novel contour representation and matching algorithm, based on local curvature scale, is proposed in this paper. First, build each point´s c-scale segment and calculate the curvature of contour points. Then, the invariant characteristic curve is established based on curvature integral, which is invariant to RST (rotation, scale and translation). Finally, the matching points of contours are captured by measuring the similarity of invariant characteristic curves. Experimental results show that this method can achieve better performance than previous methods. Also it fits for the matching between two closed contours, two open curves and the matching between an open contour and a part of closed contour. The proposed method reduces the impact of noise and scale variation effectively, and it has better robustness to rotation, scale and translation of contour.