Optik 127 (2016) 6071–6080 Contents lists available at ScienceDirect Optik j ourna l ho me pa ge: www.elsevier.de/ijleo Original research article Performance evaluation of multivariate texture descriptor for classification of timber defect Ummi Raba’ah Hashim a,∗ , Siti Zaiton Mohd Hashim b , Azah Kamilah Muda a a Computational Intelligence and Technologies Research Group, Faculty of Information and Communications Technology, Universiti Teknikal Malaysia Melaka (UTeM), Melaka, Malaysia b Soft Computing Research Group, Faculty of Computing, Universiti Teknologi Malaysia (UTM), Johor, Malaysia a r t i c l e i n f o Article history: Received 4 November 2015 Accepted 6 April 2016 Keywords: Wood material Timber surface image GLDM Texture Timber defect Feature extraction Classification performance Defect detection Automated vision inspection a b s t r a c t This paper presents performance evaluation of texture features based on orientation inde- pendent Grey Level Dependence Matrix (GLDM) for the classification of timber defects and clear wood. A series of processes including feature extraction and feature analysis were implemented to facilitate data understanding in order to construct a good feature set that could significantly discriminate between defects and clear wood classes. To further evalu- ate the discrimination capability of the features extracted, classification experiments were performed on defects and clear wood images of Meranti timber species using common classifiers. The classification performance were further compared between other timber species which are Merbau, KSK and Rubberwood. Results from the analysis reveals that the proposed texture features provide better performance than other feature sets from related works, performs acceptably well across various defect types and across multiple timber species. © 2016 Elsevier GmbH. All rights reserved. 1. Introduction An issue that presents an ongoing challenge to the timber detection problem is selecting the most appropriate features to use to differentiate clear wood and defects. Visually, defect characteristics are mostly similar but a broad range of natural variations does exist across defects, even the same types of defect. These possible variations include the size and/or shape of the defect and tonal variations. In addition, there are differences in grain appearance across different timber species, with each timber species having a unique grain appearance. In response to the timber defect detection problem, colour or tonal information alone is not enough to characterise a specific defect, for example, knots can be as dark as bark pockets and some of them could have the same colour as clear wood [1]. Although most defects appear darker than clear wood, a defect can appear as dark as the wood grain itself in some cases. Therefore, tonal properties alone are not sufficient to characterise timber defects [2]. Furthermore, since the samples acquired are of multiple species, there will be variation in timber colour. Therefore, tonal measures are definitely not suitable to represent defects across various timber species. Shapes and texture features are of equal importance to differentiate between clear wood and defects as well as types of defect [2]. But the irregularities in the shape and size of defects minimises the usefulness of shape features in addressing the timber defect detection problem as representing the various possible shapes of a defect is quite difficult. ∗ Corresponding author. E-mail address: ummi@utem.edu.my (U.R. Hashim). http://dx.doi.org/10.1016/j.ijleo.2016.04.005 0030-4026/© 2016 Elsevier GmbH. All rights reserved.