Contour-Based Similarity Retrieval of Trademarks Using Ellipse-Ring Layers

Abstract

This paper focuses on solving asymmetrical deformation problem of retrieving, and introduces a contour-based shape descriptor for retrieving both simple and complex shaped-images. In this method, we use the morphological operation to extract the boundary points from binary shape images. Then, the extracted points are partitioned into three groups according to Principal Component Analysis. We compute the statistics properties of the distribution information for each group. This method is rotation, scaling and translation invariant. Experimental results show that the proposed method is robust and efficient. In addition, this method can effectively overcome the limitation of contour-based shape descriptors.