Cell Histograms Versus Color Histograms for Image Representation and Retrieval

Abstract

This paper presents a new approach for content-based image retrieval, which is based on the well-known and widely used color histograms. Previous approaches have used a single global color histogram (GCH) for the whole image, or local color histograms (LCHs) for cells within a grid of fixed size. Our approach is also based on a grid of cells, but unlike the latter it uses a cell histogram for each of the colors actually present in the images, representing how that color is distributed among the image cells - hence the name Cell/Color Histograms. Our experiments have shown that the actual number of colors present in images is often low. Thus we are able to achieve performance comparable to using LCHs within a grid, but with a much smaller space overhead. Furthermore, the proposed approach is very flexible in the sense that the user has alternative ways to calibrate the trade-off between space overhead and retrieval effectiveness. In fact, we have been able to outperform GCHs (typically a compact representation) in terms of effectiveness requiring less storage space.