Content based image retrieval using various distance metrics

Abstract

Content based image retrieval (CBIR) provides an effective way to search the images from the databases. The feature extraction and similarity measures are the two key parameters for retrieval performance. A similarity measure plays an important role in image retrieval. This paper compares six different distance metrics such as Euclidean, Manhattan, Canberra, Bray-Curtis, Square chord, Square chi-squared distances to find the best similarity measure for image retrieval. Using pyramid structured wavelet decomposition, energy levels are calculated. These energy levels are compared by calculating distance between query image and database images using above mentioned seven different similarity metrics. A large image database from Brodatz album is used for retrieval purpose. Experimental results shows the superiority of Canberra, Bray-Curtis, Square chord, and Square Chi-squared distances over the conventional Euclidean and Manhattan distances. © 2012 Springer-Verlag.