An Efficient Content Based Image Retrieval Framework Using Machine Learning Techniques

Abstract

A Content-based image retrieval (CBIR) framework is proposed for diverse collection of images with distinct semantic categories. For effective image categorization and retrieval, the semantic category of image is considered. The low-level features (color, texture, shape and edge) are extracted and its dimensions are reduced using Principal Component Analysis (PCA). To avoid misclassification in Support Vector Machine-“Pairwise Coupling Technique” (SVM-PWC), SVM-PWC with Fuzzy C-Mean (FCM) clustering techniques and entire DB search is used. To reduce image search space, the images are prefiltered using SVM-PWC and FCM techniques. Experiments are conducted over COREL dataset consisting of 1000 images with 10 distinct semantic categories.Analysis of precision-recall for SVM-PWC and SVM-PWC with FCM clustering techniques is reported. The accuracy and testing time for SVM-PWC, SVM-PWC with FCM and Prefiltered FCM is measured. The efficiency of proposed CBIR framework is measured in the reports.