Histogram Re nement for Content-Based Image RetrievalGreg

Color histograms are widely used for content-based image retrieval. Their advantages are eeciency, and insensitivity to small changes in camera viewpoint. However, a histogram is a coarse characterization of an image, and so images with very diierent appearances can have similar histograms. We describe a technique for comparing images called histogram re-nement, which imposes additional constraints on his-togram based matching. Histogram reenement splits the pixels in a given bucket into several classes, based upon some local property. Within a given bucket, only pixels in the same class are compared. We describe a split histogram called a color coherence vector (CCV), which partitions each histogram bucket based on spatial coherence. CCV's can be computed at over 8 images per second on a standard workstation. A database with 60,000 images can be queried using CCV's in under 6 seconds. We demonstrate that histogram reene-ment can be used to distinguish images whose color histograms are indistinguishable.