Use of equalized histogram CG on statistical parameters in bins approach for CBIR

This paper explores the novel and simple technique for feature extraction for CBIR. It works for improving the retrieval accuracy of the CBIR system along with reduction in the feature vector dimension. Feature extraction process is based on the bins approach. Image information is extracted to the eight bins formed by partitioning the histogram using CG (center of gravity). The image is separated into R, G and B planes. For each plane original and equalized histograms are calculated. Histograms are partitioned in two parts and image contents are segregated into 8 bins. The image contents extracted to 8 bins are represented using the first four statistical moments. Feature vector databases are prepared for all four moments. All feature vector databases are tested using the same set of query images fired on them. Query and database image feature vectors are then compared using three similarity measures namely Euclidean distance ED, Absolute distance AD and Cosine correlation distance CD. Results obtained are evaluated using three parameters PRCP i.e Precision Recall Cross over Point, LS: Longest String and LSRR Length of string to retrieve all relevant images from database. The system proposed in this paper is experimented with database of 2000 BMP images containing 20 different classes from various sources including Wang database. Each class contains 100 images of its own category.