Image classification and retrieval is an important area of pattern recognition. In this paper, we propose a new algorithm to describe the shape of image based on regional boundaries. A shape group is composed of some separate shapes. For example, trademarks, traffic signs and so on are all shape groups,We propose a more accurate method to classify and retrieve shape group database.The results of this paper are as following:First, we make a shape group database which contains1660shape groups. This database is divided into83classes and each class is formed by20shape groups that are composed by shapes from the known database MPEG-7-PartB.Second, we propose a new characteristics expression algorithm based on V-system of degree1and use this algorithm for shape groups classification and retrieval. The detail of this algorithm is explained as following:A shape group is accurately expressed with finite basis functions of V-system, and then its feature vector is extracted in the frequency domain. With features vector obtained, similarity measure value between two shape groups is calculated and then a new shape group classification and retrieval method is proposed. As the V-system can accurately reconstruct the shape group, the features of shape group obtained by V-system is accurate and reliable. The classification and retrieval experiment results show the superiority of our method over the classical Fourier descriptor, Zernike moments, invariant moments, and geometric center moments. |