Image retrieval based on ameliorated color histogram: Sector model
Taijun Li,YaoHong Kang,Xiaobin Wen,XueSong Pan
2006-01-01
Journal of Computational Information Systems
Abstract:Color histograms are the color features which are widely adopted in a lot of image retrieval systems. Description of the color histogram is the proportion of different colors in the whole image. It is not concerned about the spacial distributing information of colors. In order to rectify this defect and improve the resolution of colors, it is necessary to calculate color histograms thereby partitioning the color space into a lot of small bins. However, the amount of calculating burden increases in this way and may be very enormous when the number of bins is large, that is infaust to set indexes in a large scale image database. According to ameliorating color histograms, a new image retrieval algorithm based on image projection is presented in this paper. Namely, by ameliorating the HSV color model, the hue, saturation and intensity is projected respectively onto the radial and angle directions in a polar coordinate system, thus six projective histograms are obtained. The projective histograms of hue and saturation characterize the color distribution in space, and the projective histogram of intensity reflects the shape of the image to a certain extent. Then, take the first three central moments (mean, variance, skewness) of these six projective histograms as features to compute the distances between images. Experimental results show that this algorithm performs tersely and quickly and has better recall rate and precision than the global color histograms.