Wood Identification Based on PCA, 2DPCA and (2D)2PCA

In this paper, a novel wood identification approach based on the two-directional two-dimensional PCA ((2D)2PCA) method is proposed in contrast to the principal component analysis(PCA), two-dimensional PCA(2DPCA), column-directional 2DPCA(c2DPCA). PCA is a classical technique used to find patterns in high dimensional data. Wood identification based on PCA must transform 2D image matrix into 1D vector, and then calculate principal components from these 1D vectors. 2DPCA, c2DPCA and (2D)2PCA methods are based on 2D image matrix as opposed to classical PCA. All these wood identification methods involve seven identical steps: (1) calculating sample's mean; (2) demeaning all images; (3) calculating the demeaned sample's covariance matrix; (4) eigenvalues and eigenvectors decomposition of covariance matrix; (5) eigenvectors selection according to the largest eigenvalues to construct feature space; (6) extracting features by projecting image onto feature space; (7) classifying by the nearest neighbor classifier with Euclidean distance in feature space. Experiments using PCA, 2DPCA, c2DPCA, (2D)2PCA methods are involved in this paper. By comparing PCA, 2DPCA and c2DPCA in these experiments, it is revealed that (2D)2PCA is a more efficient method in wood identification.