Study of relative effectiveness of features in content-based image retrieval

Content-based retrieval in image databases requires appropriate features of images to be derived and used in the indexing and searching process. Fast and accurate retrieval is crucial from a user point of view. The kinds of features used, their organization in suitable data structures, and the similarity search scheme, directly affect the speed and quality of content-based retrieval. In this paper, we evaluate the relative significance of three different image features - geometry, moments, and Fourier descriptors in the context of content-based retrieval, and present experimental results. The evaluation metrics are retrieval quality and search time. These could be used to tune the search process which speeds up retrieval without significantly affecting retrieval quality.