An Effective Approach towards Contour-Shape Retrieval

Distance histogram has been used for shape representation and retrieval. Though it is efficient and effective, it looses the spatial information of a contour. In this paper, the pixels of a contour are classified into different types, convex, concave and smooth pixels. In addition to distance histograms, two novel descriptors named spatial location distribution (SLD) and spatial distribution entropy (SDE) are introduced. These descriptors have powerful descriptive power for contour with more spatial information. Comparisons are conducted between our method and several other feature descriptors. The results show that the new method is efficient and it provides noticeable improvement to the performance of shape retrieval.