Performance analysis of hybrid Evolutionary computational Techniques f or image feature Selection and texture Classification

Abstract

Feature selection is the process of selecting the most efficient feature subset from the available large volume of features. Any classification problem requires a set of features to properly identify the class of a specified data point. The process of recognizing the relevant and efficient features that identifies a class uniquely is called Feature Selection and this research work deals with feature selection of texture features for efficient texture classification. Feature subsets can be evaluated through two methods of subset selection. Individual feature can be tested for its efficiency by using only that feature as the classification base. Random feature subset selection aims to combine more features randomly and the efficiency of that specific group of features will be evaluated through standard classifiers. Evolutionary Algorithms used in feature selection through random combination of features to form new subsets guided by the process of Bio-inspired computation. Fast computation, parallel evaluation of many solutions, better subset selection is some of the reasons for choosing Evolutionary Algorithms for Feature Selection. newlineThe first module involves a hybrid evolutionary algorithm designed from improvised Genetic Algorithm and Tabu Search for feature selection. The features used for the primary data of Texture images are Haralick Texture Descriptors derived from Grey Level Co-occurrence Matrix (GLCM) with different distance and angle parameters. The algorithm was tested on six Standard Texture Datasets (UMD, KTH TIPS, UIUC, Kylberg, SIPI, Outex T10) and one real-time dataset (Texture Feature Selection on Fabric Data). The evaluation metrics used were accuracy, precision, recall, fitness and execution time. newline