Distinctive parts for shape classification

Shape is an important cue for object recognition, While previous research on shape classification uses all parts of a feature set (e.g. contour segments, skeleton paths), we define distinctive parts that tell an object from objects of a different type. Our approach to analyze distinctive part is based on performing a shape-based search using each part as a query into a database. Distinctive parts of a shape are consistent with objects of the same type and different from objects of other types. Experimental results demonstrate that the concept agrees with human common sense, and show the effectiveness and efficiency of our method for shape classification as well.