Performance improvement using average query fired to bins of four statistical moments for CBIR

Abstract

This paper explains the effectiveness of average feature vector used as compared to a single query image feature vector to be fired to the CBIR designed using bins approach based on the partitioning of the equalized histograms of R, G and B planes of images. The feature vectors of dimension 27 are extracted into bins holding the statistical information of first 4 centralize absolute moments of R, G and B colors separately. Three different similarity measures are used in this paper for comparing the query image and database images namely Absolute distance, Euclidean distance and Cosine correlation distance. Experimentation of this approach is demonstrated for image database of 2000 BMP images containing 100 images from 20 different classes. Three parameters are used namely PRCP, LSRR and Longest String to evaluate the performance of the approaches used in this paper for CBIR.