Abstract: Previous histogram similarity measures have disadvantages when several kinds of deformation happen to the histograms. A novel similarity measure is proposed based on average translation of histogram bins. The value differences of both corresponding and cross bins are calculated to get the average difference against bin distances, and the average translation of bins is used to measure the histogram similarity. Experiments on intensity histograms of images influenced by noise and illumination are carried out to compare the performance of different kinds of similarity measures. The results show that average translation of bins as a histogram similarity measure outperforms other measures under the condition of histogram deformation. It consists with the other measures under general conditions. Preliminary studies on the measure’s usefulness in image retrieval and object searching indicate its potential applications based on image histograms.

Key words: histogram, histogram similarity, average translation of bins, image retrieval