최대/최소 칼라 코렐로그램을 이용한 영상검색 시스템

Abstract

Color is widely used to represent an image for realizing content-based image retrieval (CBIR). Color histogram and color correlogram are the most popular techniques in the literature for color feature detection in CBIR. The color composition of an image, which is usually represented as a histogram of intensity values, is a global property which does not require knowledge of the component objects of an image. The color correlogram was proposed to characterize not only the color distribution of pixels, but also the spatial correlation of pairs of colors. Color correlogram requires a lot of storage and a large number of computations depending on the number of quantized colors used. In this paper, we tried to reduce the number of quantized colors for color correlogram. The color of a pixel is presented as RGB. The image is divided into small sub-images. The dominant color components r/g/b of the sub-images are determined. The color correlogram matrix is defined based on the maximum/minimum of color component of a sub-image for different spatial distances. The proposed algorithm has shown good performance as compared to the previous methods with reduced number of featuers.