Classification of Wood Surface Defects with Hybrid Usage of Statistical and Textural Features

Machine vision is applied to detect wood knots and cracks, to classify strong and stable woods. In order to obtain effective and efficient classification a well-defined pattern recognition and feature extraction algorithms are essential. In this paper we examine three different methods for feature extraction; Gray level co-occurrence matrix (GLCM), Local binary patterns (LBP), and statistical moments. The hybrid usage of these methods is considered. Principal Components Analysis (PCA) and Linear Discriminate Analysis (LDA) are utilized to reduce the feature vector dimension. We use Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) for classification. The classifiers are applied for five different wood knot species. The efficiency of the proposed method using hybrid features called, GLCM+LBF, GLCM+statistical moments and LBF+Statitical moments are investigated through simulations. Comparison with latest works is accomplished to show the capability of the proposed method.