“…We compared 11 similarity measures including L 1 , L 2 , L 1 (Theodoridis & Koutroumbas, 1999), angular (Androutsos, Plataniotis, & Venetsanopoulos, 1999), Canberra (Androutsos et al, 1999), Czekanowski (Theodoridis & Koutroumbas, 1999), v 2 (Antani et al, 2000), inner product (Theodoridis & Koutroumbas, 1999), Fu (Fu, Yang, Braylan, & Benson, 1993), Mahalanobis (Rubner et al, 2001), and weighted-mean-variance, WMV (Manjunath & Ma, 1996), for color histogram, edge histogram and Gabor features on the basis of classification rate [Nezam05a]. Among these measures, the v 2 was the best for color and Gabor features, while Euclidian distance performed better for edge histogram (Fig.…”