Shape Retrieval Using Eigen and Fisher Barycenter Contour

Abstract

To achieve a good performance for shape retrieval, it requires both shape representation and classifier. In this paper, the algorithm for shape matching and retrieval is developed by using Eigen Barycenter Contour (EBcC) and Fisher Barycenter Contour (FBcC). In our algorithm, the Signed Enclosed Area (SEA) signature (formed by two adjacent points of contour and its center point), computed at each scale level of Barycenter contour (BcC), is utilized as the shape representation. The BcC technique is robust to moderate amount of noise and occlusion. Furthermore, the SEA signature is invariant to general affine transformation including translation, rotation, scale and shear. Because of high dimension of the shape representation, thus, in the matching step, two classifiers have been studied. The first classifier, Eigen face technique, is employed for dimensionality reduction while the second classifier, Fisher face technique, is used for reducing dimension as well and making discrimination. Then, the similarity among shapes is measured by the normalized cross correlation (NCC). The performance of our technique is evaluated onto the affine shape database and two well-known databases, the MPEG-7 shape database part B and the Kimia's database. The experimental results illustrate that our approach gives very high retrieval efficiency over all published methods.