An algorithm for binary contour objects representation and recognition

Abstract

Representation of an object in image for machine learning applications (recognition, retrieval, identification, etc.) has to be based on a previously chosen feature. Binary shape is a very popular and commendable one. It has many advantages and can be successfully used in many applications, especially in engineering. To achieve better characteristics, various shape transformations are used. Obviously, they should be robust to as many shape deformations as it is possible. In this paper results of exhaustive exploration of a new method are presented. This method is based on transformation from Cartesian to polar coordinates, but it overcomes few problems, that were not solved yet. Above all, the proposed transform is more robust to occlusion and noise, two the most challenging problems. © 2008 Springer-Verlag Berlin Heidelberg.