Currently, high-resolution 3D morphological data can be acquired easily and cost-effectively by using CT scanners, LiDAR, and photogrammetric techniques. A pipeline using Structure from Motion (SfM) and Multi-View Stereo (MVS), which is a promising technique to reconstruct a 3D surface as point cloud data from a series of 2D images taken from different angles, has been implemented in several libraries and software products. On the other hand, we need some model to extract phenotypic values that are biologically significant from these 3D data. For describing morphological properties of plants, several types of models have been proposed from simple models such as LAI to complex ones such as functional structural plant model (FSPM). However, we have no adequate method to evaluate the differences among 3D structures of plants in terms of morphometrics because it is difficult to define the homology among them. For example, it is not obvious how to correspond a particular leaf of a tree to which leaf of another tree. In this study, I propose a model or a representation, which can provide just enough information to both represent and reconstruct plant styles. Plant style is aboveground 3D spacial structure of plants which is composed of several organs hierarchically. Here, I adopted a multi-variable probability distribution of position, area, direction, shape of leaves for representing plant styles as foliage. By using this model, a plant style can be not only represented as a multi-variable probability distribution but also compared with another one on a statistical manifold. Moreover, virtual individual plants and canopies can be generated based on the distribution. It implies that the model was designed as features that have enough resolution to reconstruct plant styles.
