Comparing the influence of color spaces and metrics in content-based image retrieval

Visual information retrieval systems have been developed from the necessity of searching and recovering data in libraries with million of daily produced digital images. In image retrieval based on their color content, the distance between color histograms defines usually the matching degree. Results of an image querying depend, at least, on the following features: the underlying color space, the number of histogram bins, and the metric used for histogram matching. Several systems have been designed for image retrieval. Comparison among them is not feasible because they use completely different implementation aspects. The purpose of this paper is to evaluate the influence of color spaces and metrics on the performances of recovering by color similarity based on color histograms. A system that performs queries through the Internet and allows one to combine metrics and color spaces is implemented. We design a database with groups of perceptually similar color image, with parameters to measure the performance among combined approaches, and with a user-consult simulator. This leads to a process of establishing the influence of spaces and metrics which evaluate the better combination on retrieval