Fast signature-based color-spatial image retrieval

Abstract

This paper presents a novel method of retrieving images based on color-spatial information. For each image, its color-spatial information is compactly captured by a set of color signatures and an image signature. During the image retrieval process, the color and image signatures of the query image and those of the database images are compared to determine the set of candidate images to be retrieved. In particular our similarity measure takes into account contributions of colors that are perceptually similar and the weightage on the object colors. To speed up the retrieval process, we also propose a filtering mechanism that employs a simple similarity measure to prune away irrelevant images quickly, before computing the more complex similarity measure on potential candidate images. We implemented the proposed method and conducted an experimental study on an image database consisting of 12,000 images. Our results show the retrieval effectiveness and efficiency of the proposed color-spatial approach.