Efficient, Compact, and Dominant Color Correlogram Descriptors for Content-based Image Retrieval

摘要

Color is one of the most important and widely used cues in content analysis and retrieval. However, most promising color descriptors consume massive amounts of computation and storage, which is a serious drawback. One of these promising color techniques in image retrieval is the color correlogram, but the technique also suffers from the aforementioned drawbacks. In this paper, we present two compact representations of the color correlogram. The first representation is the compact-generalized correlogram, which compresses colors and generalizes the distances of the original correlogram descriptor. The second representation is the dominant color-based correlogram, which is also a compact and conceptual correlogram descriptor. This representation computes the spatial correlations of the dominant colors of a few images instead of a large number of quantized colors used by the original descriptor. The two representations are integrated. The experimental results prove the high effectiveness and feasibility of the proposed descriptors through two large image databases (i.e., Corel-10K and Cartoon-11K) using ARR, ANMRR, P(10), and MAP metrics.