A semiautomatic color space quantization method is proposed, which models the human perception of color names, and is suited for image retrieval application. Comparison of the proposed Automatic Classification step with the clustering and the manual quantization methods shows 1.5-8.8% increase in the perceptual quality, and 1.7-1,327.0 times decrease in the computational cost. The proposed Manual Reclassification step additionally brings over 30% increase in the perceptual quality. Applied to retrieval from the database of 1,100 color images and compared with the manual quantization method, the proposed method leads to 16% increase in the retrieval precision, enabling a more effective retrieval from large scale image databases.
展开▼