Shape Based Image Classification and Retrieval with Multiresolution Enhanced Orthogonal Polynomial Model

This paper proposes a simple edge-based shape representation with multiresolution enhanced orthogonal polynomial model and morphological operations for effective classification and retrieval. The proposed method consists of four phases: (1) orthogonal polynomial computation, (2) edge image construction, (3) approximate shape boundary extraction with morphological operation, and (4) invariant Hu’s moment computation. Initially, the orthogonal polynomials are computed and the obtained coefficients are reordered into one-level subband-like structure. Then, the edge image is obtained by utilizing gradient in horizontal and vertical directions from the detailed subbands of the reordered structure. The rough shape boundary is computed with morphological operation. The global invariant shape features such as Hu’s moment and eccentricity are extracted in the fourth phase. The obtained features are termed as global shape feature vector and are used for retrieving and classifying similar images with Canberra distance metric and Bayesian classification, respectively. The efficiency of the proposed method is experimented on a subset of standard Corel, Yale, and MPEG-7 databases. The results of the proposed method are compared with those of existing techniques, and the proposed method provides significant results.