Comparative Performance of Linear and CG Based Partitioning Of Histogram for Bins Formation in CBIR

This paper presents the CBIR based on bins approach. It introduces a new idea of partitioning the histogram into three parts using Centre of gravity. This partitioning leads to generation of 27 bins. In this work we have tried to reduce the feature vector dimension to just 27 bins out of 256 histogram bins. This paper elaborates the bins approach using linear (LP) and centre of gravity (CG) based histogram partitioning for generation of 27 bins. Image contents extracted to these bins are the count of pixels falling in the specific range of intensities plotted in the R, G and B histograms. These contents are process further by computing the statistical first four moments Mean, Standard deviation, skewness and kurtosis. The moments are computed separately for R, G and B intensities and treated as separate feature vectors and stored in separate feature databases. Experimentation work is carried out using database of 2000 BMP images having 20 classes including few from Wang database. Core part of this CBIR i.e comparison of query and database images is facilitated using three similarity measures namely Euclidean distance(ED), Absolute distance (AD) and Cosine correlation distance (CD). Performance of the proposed CBIR system is evaluated using three parameters Precision Recall Cross over Point (PRCP), Longest String (LS), Length of String to Retrieve all Relevant (LSRR).