Refine
Document Type
Keywords
Institute
There are many different approaches to measure dissimilarities between images on the basis of color
histograms. Some of them operate fast but generate results in contradiction to human perception. Others
yield better results, especially the Earth Mover’s Distance (EMD) (Rubner et al., Int J Comput Vis,
40(2):99–121, 2000), but its computational complexity prevents its usage in large databases (Ling et
al., IEEE Trans Pattern Anal Mach Intell, 29(5):840–853, 2007). This paper presents a new intuitive
intelligible approximation of EMD. The empirical study tries to answer the question whether the good
results of EMD justify its long computation time. We tested several distances with images that were
changed by normally-distributed failures and evaluate their results by means of the adjusted Rand index
(Hubert et al., J Classif, 2:193–218, 1985).