Shape Matching and Recognition using Hybrid Features from Skeleton and Boundary

This paper presents a novel approach for effective matching of similar shapes from skeleton and boundary features. The features identified from the shape are the junction points, end points, and maximum length from single pixel pruned skeleton of the shape. Another two features identified from the boundary are junctions and boundary length of the shape. These five features are then used for shape matching. We tested these features on Kimia shapes dataset and tools dataset. The matching process from these features has produced good results, showing the probable of the developed method in a variety of computer vision and pattern recognition domains. The results demonstrate these features are rotational and transform invariant.