A Novel Machine-Learning Framework-based on LBP and GLCM Approaches for CBIR System

Abstract

This paper presents a Multiple-features extraction and reduction-based approaches for Content-Based Image Retrieval (CBIR). Discrete Wavelet Transforms (DWT) on colored channels is used to decompose the image at multiple stages. The Gray Level Co-occurrence Matrix (GLCM) concept is used to extract statistical characteristics for texture image classification. The definition of shared knowledge is used to classify the most common features for all COREL dataset groups. These are also fed into a feature selector based on the particle swarm optimization which reduces the number of features that can be used during the classification stage. Three classifiers, called the Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Decision Tree (DT), are trained and tested, in which SVM give high classification accuracy and precise rates. In several of the COREL dataset types, experimental findings have demonstrated above 94 percent precision and 0.80 to 0.90 precision values.