Transactions on Machine Learning and Data Mining (ISSN: 1865-6781)

Volume 5 - Number 1 - July 2012 - Pages 45-62

Recognition of Wood Porosity Based on Direction Insensitive Feature Sets

S. Pan1,2, M. Kudo1

1 Graduate School of Information Science and Technology, Hokkaido University, Sapporo, 060-0814, Japan
2 Department of Electronic Business, Hefei University of Technology, Hefei, 230009, China


The size and configuration of pores are key features for au- tomatic wood identification. In this paper, two feature sets insensitive to rotation and transition change are extracted and then used for construc- tion of decision trees for recognizing three different kinds of pore distribu- tions in wood microscopic images. The contribution of this paper lies in three aspects. Firstly, two direction insensitive sets of features for poros- ity classification are designed and extracted, Secondly, for turning the found classification rule into human-readable knowledge, decision trees are built with the feature sets by C4.5 algorithm; Finally, rules extracted from the decision trees are explained according to domain knowledge of wood science.

PDFDownload Paper (298 KB)

Back to Table of Contents