An efficient shape recognition scheme based on three new ideas is presented. First, a new method is presented for estimating real-valued boundary points from the integer-valued coordinates and the grey-level values of binary images. Secondly, we can quickly determine all of the feature points with this representation by using an angle calculation formula and the separating technique which are also proposed in this paper. Then, each shape is represented by an array of pairs, where each pair contains the coordinate of a feature point and its distance from the centroid. Thirdly, in the matching process, we also propose a new split-merge technique to assist in the shape matching. The effectiveness of the shape recognition scheme is clearly proven by the good recognition rates of our experiments.