Global warp metric distance: boosting content-based image retrieval through histograms

Abstract

This work presents a new distance function - the global warp metric distance - to compare histograms used as a feature to index image databases in content based image retrieval environments. The metric histogram represents a compact, but efficient alternative to the use of traditional gray level histograms to represent images. The global warp metric distance (GWMD) enhances the comparison between histograms, replacing the rigid bin to bin evaluation by the warp method, which allows a local "adjustment" of one histogram to the other during the distance calculation, introducing a global matching of the curves. Besides this, GWMD applies a set of geometric global features of histograms to determine the final distance. Results on similarity retrieval in medical images demonstrate the superiority of the proposed approach in analyzing image sets that present brightness and contrast disparities: it reduces the amount of both false positive and false negative retrievals. Moreover, these results comply with similarity evaluations performed by domain specialists.