Shape Recognition Using Segmenting and String Matching

Abstract

This paper presents an efficient way to represent objects. The image of the object is converted into an edge image. Important points of the curve are identified by the dominant point detection method. A line segment of every two consecutive important points is a categorical line segment or a non-linear line segment. Nonlinear segments are fitted as circular arcs. In addition, the compactness of approximate polygons is used as a feature in the shape recognition process. Experimental results show that using this new global feature has better recognition performance than traditional features such as relative distance, length and angle. Overall the new method is efficient and effective in representing and recognizing shapes.