Shape Recognition Using a New Spatial Representation and a D.P. Matching Algorithm

We propose a new method for the recognition and retrieval of shapes whose contours are simple closed curves. First, we give a new shape representation by a sequence of 2D-vectors of angles, which is independent of rotation and scaling. Each 2D-vector captures the local shape information around a point in the contour. Next, we apply a dynamic programming method to match the points of two contours and identify the outlier (unmatched) points in each contour with respect to the other. Then, we define the similarity of a pair of contours by the number of unmatched points and the matching cost for the matched points. Finally, the recognition and retrieval are done using the nearest-neighbor method. Our experiments on MPEG-7 database show that the performance of the new algorithm is very similar to that of the best-known algorithms in the literature in spite of its significantly less computational complexity and simplicity.