Comparison of similarity measures in HSV quantization for CBIR

Abstract

Researchers implemented various similarity measure for CBIR using HSV Quantization. Implemented similarity measures on this study is Euclidean Distance, Cramer-von Mises Divergence, Manhattan Distance, Cosine Similarity, Chi-Square Dissimilarity, Jeffrey Divergence, Pearson Correlation Coefficient, and Mahalanobis Distance. The purpose of study is to measure the performance of image retrieval of the CBIR system using HSV Quantization for each of the similarity measures. The performance of similarity measures are evaluated based on precision, recall, and F-measure value that obtained from test results performed on the Wang dataset. Similarity measures were performed on each of the categories (Africa, Beaches, Building, Bus, Dinosaur, Elephant, Flower, Horses, Mountain, and Food) that has 100 images of each its category. The test results showed that the highest precision valued are 100% provided with Jeffrey Divergence on Dinosaur category. The best average precision value of all categories is provided with Jeffrey Divergence, i.e. 87.298%. In generally, the best average precision value is Dinosaur category (Euclidean Distance, Manhattan Distance, Cosine Similarity, Chi-Square Dissimilarity, Jeffrey Divergence, and Pearson Correlation Coefficient). The next of average precision value is on Flower category for Cramer-von Mises Divergence, and the last category is on Bus category that provided with Mahalanobis Distance. The highest average recall valued is 92% on Horses category that established to Cosine Similarity. The best average recall valued for all categories is on Manhattan Distance, i.e. 38.700%. In generally, the best average recall valued is on Horses category that provided with Cramer-von Mises Divergence, Manhattan Distance, Cosine Similarity, Chi-Square Dissimilarity, Jeffrey Divergence, Pearson Correlation Coefficient, and Mahalanobis Distance. The best average recall value of the Euclidean Distance is Africa category. The highest F-measure value is 87.255% on Horses category provided with Cosine Similarity. The experiment result showed that the highest F-measure valued is always on Horses category. The highest F-measure value in general provided with Manhattan Distance (Africa, Beaches, Building, Bus, Dinosaur, Elephant, Flower, Mountain, and Food), while the highest F-measure valued of Horses category provided with Cosine Similarity.