An optimized feature selection CBIR technique using ANN
I. Thusnavis Bella Mary, A. Vinotha Vasuki, M. A. P. Manimekalai · 2017 International Conference on Electrical, Electronics, Communication, Computer, and Optimization Techniques (ICEECCOT) · 2017
This paper focus in image retrieval technologies that allow users to retrieve images using a large number of highly selective features. There exist two main challenges. Firstly, high retrieval efficiency. Secondly, reduce dimensionality of image feature space which leads to less computation time. In this work, we developed a novel approach that attempts to address these two challenges. Unique point of this approach is that integration of hybrid technique, feature selection technique and artificial neural network. In this paper, hybrid descriptors (i.e.) a combination of three features namely color, texture and shape are extracted, from which optimal descriptors are selected using dimensionality reduction genetic algorithm. The Back Propagation Neural Network (BPNN) is used as a classifier. The idea behind BPNN is its high accuracy, error minimization and flexibility. The experimental results demonstrate our method that outperforms in terms of accuracy and retrieval time compared to the other existing methods. Also the various performance metrics have been discussed and from the experimental results, it was found that Mean Normalized Retrieval Order (MNRO) is better than the other performance measures.