Classification of wood micrographs by image segmentation

Abstract

The principal aim of this study is to classify wood species using scanning electron microscopy (SEM) micrographs obtained with 1500 × magnification and processed by image segmentation. The results show that it is possible to observe differences among species in the wood texture at this magnification. The micrographs have been processed in a simple way using segmentation and object recognition to identify the cross-section tracheids belonging to earlywood of 7 different timber species: Fagus sylvatica, Castanea sativa, Juglans regia, Eucalyptus globulus, Hymenaea courbaril, Pinus silvestris and Pinus radiata. We have analyzed the shape, number and distribution of the tracheids using 5 features: circularity, rectangularity, number of tracheids, distance between tracheids and average area. The extracted features are classified using different statistical methods: Linear Discriminant Analysis (LDA), Quadratic classification, Logistic regression, K Nearest Neighbors (KNN), Naïve Bayes, Support Vector Machines (SVM) and Neural Networks. A comparative study using gray level co-occurrence based features is also presented, with the improvement of using the segmentation method. Moreover, some additional results showing the possibility of using fractal analysis in this framework complete the research.