Color Space Quantization for Color-Content-Based Query Systems

Abstract

Color histograms are widely used in most of color content-based image retrieval systems to represent color content. However, the high dimensionality of a color histogram hinders efficient indexing and matching. To reduce histogram dimension with the least loss in color content, color space quantization is indispensable. This paper highlights and emphasizes the importance and the objectives of color space quantization. The color conservation property is examined by investigating and comparing different clustering techniques in perceptually uniform color spaces and for different images. For studying color spaces, perceptually uniform spaces, such as the Mathematical Transformation to Munsell system (MTM) and the C.I.E. Laaaba, are investigated. For evaluating quantization approaches, the uniform quantization, the hierarchical clustering, and the Color-Naming-System (CNS) supervised clustering are studied. For analyzing color loss, the error bound, the quantized error in color space conversion, and the average quantized error of 400 color images are explored. A color-content-based image retrieval application is shown to demonstrate the differences when applying these clustering techniques. Our simulation results suggest that good quantization techniques lead to more effective retrieval.