Monochromatic textures' features extraction using extended GLCM approach for classification of autonomous cleaning robot work area

Abstract

One of commonly used methods for scanning the work area around mobile robot is to use machine vision. This paper's focus is on extracting features from monochrome natural textures for the purpose of texture classification using extended Gray Level Coincidence Matrix (GLCM) approach. Main idea of this approach is to slice original image into smaller parts, calculate four well-known Haralick's Features for each part separately and then use one of commonly used statistical measures to obtain series of features for task of classification. Simulations using texture base derived from popular Amsterdam Library of Textures (ALOT) database were performed. Evaluation of classification performance with this extended method for different number of slices was performed using Re-substitution Loss, F-measure and Cross-validation loss of calculated classifiers as quality criteria. In general, obtained results show that it is possible to improve classification quality by introducing this extended approach.