Texture representations using subspace embeddings
Pattern Recognition Letters
| Hi-index | 0.00 |
Wood classification is fairly important to the forestry industry and forest conservation. There have already been some microcomputer-assisted classification methods which accomplish the classification mainly with feature image analysis. But it is also necessary to find out a more effective method to optimize the classification process. So we propose to apply Principal Component Analysis (PCA), 2-Dimensional Principal Component Analysis (2DPCA), 2-Dimensional 2-Dimensional Principal Component Analysis ((2D)2PCA) and Linear Discriminant Analysis (LDA) to wood texture feature extraction and expect to obtain a better effect. With two sets of experiments, we verify that not only PCA and its corrective methods but also LDA are completely feasible in extracting wood texture features and LDA is a little better than PCA in the cross section sample classification but not good in the tangential section sample classification. Therefore, the cross section is better than the tangential section to act as the experiment sample in wood classification. But according to our experimental results, LDA does not outperform 2DPCA and (2D)2PCA because of lack of training samples, which prove that the latter are more preferred algorithms to identify wood.