Table 1.
Wood species used in the present study.
Fig 1.
Schematic diagram of the computational procedure.
Fig 2.
Effect of resolution on image analysis.
(a) Changes in the accuracy of classification with decreasing image size. The error bar shows the standard deviation. (b) Examples of images for various image sizes corresponding to the arrows in (a). The images are an enlarged part of an image of C. crenata.
Fig 3.
Hierarchical clustering of all the images using the features of (a) SIFT-LDA and (b) CC-LDA. The labels show the dominant taxon groups in the clusters. The numbers in parentheses indicate the number of dominant taxon groups per total number of groups in a cluster.
Fig 4.
Comparison of the compositional ratio of the features detected by the keypoints.
The keypoints in the images of seven species are clustered into 18 groups by k-means. (a) Hierarchical clustering of the centroids of 18 groups. The 18 groups were further divided into five subgroups. (b) Composition ratios of the 18 groups for each species. The numbers on the right indicate the average number of keypoints per image. (c) An enlarged part of an image of Q. crispula with the keypoints classified in each subgroup. The images show that each subgroup roughly corresponds to some shape feature and anatomical element.
Fig 5.
Taxon-specific features based on keypoint clustering.
(a) Keypoints included in the clusters that occurred more often in both Q. acutissima and Q. phillyraeoides than in other species. (b) Keypoints in the cluster that occurred significantly more in L. edulis than in other species.
Fig 6.
Comparison in the distribution of pore size.
The data reported for each species are the averages of all images. The bars below indicate the size of various cell lumens.