A New Integrated Method for Shape Based Image Retrieval

This paper proposes a novel and efficient shape retrieval scheme, which is robust to RST (Rotation, Scaling, and Translation). The proposed approach integrates global and local shape descriptors for accurate retrieval, where the global descriptors are obtained from the Fourier transformation, and the local descriptors are obtained from a one-level wavelet transformation. The global and local similarity scores for each query and candidate image are individually computed using different measures. A Gaussian-based fuzzy method then combines the global and local similarity scores into one similarity membership, which measures the overall similarity between the query and candidate images. Experiments are performed on four different databases and the retrieval results demonstrate that the proposed approach not only accurately and efficiently retrieves shapes but also outperforms some peer systems in the literature.