Relative Examination of Texture Feature Extraction Techniques in Image Retrieval Systems by Employing Neural Network: An Experimental Review

Abstract

In this paper, the performance of the four most significant texture feature extraction techniques of content-based image retrieval (CBIR) systems are technically analyzed and compared in two phases. In the first phase, simply texture features are extracted by various techniques on three important CBIR datasets which are Wang, Corel-5K and Corel-10K and retrieval parameters are evaluated. In the next phase, artificial neural network, i.e., cascade forward back propagation neural network (CFBPNN) is employed as a classifier after the step of feature extraction which increases the accuracy and average precision of the system. The main issue with CBIR systems is the proper selection of techniques for the extraction of low-level features of the images which comprises of color, texture, and edge. Among these features, texture is one of the most influential features. This selection of features completely depends upon the type of images to be retrieved from the database. The four important texture techniques i.e. Grey level co-occurrence matrix (GLCM), discrete wavelet transforms (DWT), Gabor transform and local binary pattern (LBP) are experimented here and performance parameters are evaluated. So this analysis will help the researchers in selection of proper texture feature techniques depending upon the type of images for CBIR systems. During phase 1, which is without classifier the LBP technique has the highest precision which is around 82, 70 and 66% on Wang, Corel-5K, and Corel-10K respectively. The results spectacles a significant improvement with the use of CFBPNN and these results now comes up to around 94, 83, and 79%, respectively, on the same datasets.