A modified structural method for shape recognition

Abstract

Shape recognition plays an important role in machine vision applications. The effective recognition algorithm should be less complicated and more accurate. In this paper, a structural descriptor for shape recognition is presented which has less complexity and more accuracy comparing with existing algorithms. Obviously, the boundary of shape is an important property in shape representation and description. In proposed method, the centroid of shape is considered as a basic point. Then, two structural properties such as distance and angles between the centroid and shape contour are calculated. Finally, by combining these two structural features, a new Feature-Space is constructed. The proposed shape descriptor can measure periodical, smoothness and symmetry of shapes and can be used as a modified method for shape recognition.