Performance Comparison of Feature Weighting on Content Based Image Retrieval Using Bayesian Network and Genetic Algorithm

Abstract

This research proposes performance comparison on image retrieval using bayesian network and genetic algorithm, which combines color, shape and texture information. Histogram-162 is used for color feature, edge direction histogram for shape feature and co-occurrence matrix for texture feature. Combining multiple features in image retrieval process can be implemented using feature weight assignment. Bayesian network and genetic algorithm is used to find the optimal weight. This research used 1050 images with various classes i.e car, lion, sunset, texture, bear, elephant, arrow, landscape, reptile and aircraft. Experiment results shows that genetic algorithm has better precision than bayesian network on recall between 0.0 and 0.3.