Development of an appearance texture generation system for single plant organ image

Abstract: This paper presents a plant organ appearance textures generation system using a single image to improve the efficiency of three-dimensional digital media content production and simplify the tedious process of producing the textured appearance of plant organs in images. The system consists of six modules: diffuse reflection texture, transmittance texture, specular textures, normal map, ambient occlusion (AO) map and real-time visualization. The system takes a single organ texture as input. Firstly, the input texture is decomposed into diffuse reflection texture and shading texture by the diffuse reflection texture module. The two textures thus obtained are also used as input images for other modules. Diffuse reflection texture module adopts an energy constrained intrinsic decomposition method, which is stable and quick. The transmittance texture module can transform a diffuse reflection texture into a transmittance texture. The core of the module is a transmittance empirical model based on the PROSPECT model, which is a widely used leaf-scale radiation transfer model. To obtain this empirical model, we first generate the spectral data of leaf reflectance and transmittance under different parameters by PROSPECT model and transform them into RGB color space. Then we invert the statistical formula of plant leaf transmittance and reflection by using weighted least squares regression. The formula yields the transmittance color by diffuse reflection color. Specular intensity and roughness parameter describe the directional characteristics of leaf surface reflection. In theory, these two parameters can't be accurately calculated from a single image. Therefore, specular textures module focuses on providing users with an interactive tool to speed up the generation of textures. We assume that the materials of leaf-blade regions that show similar diffuse reflectance are similar, as are their specular parameters. Therefore, in the specular textures module, the user interactively selects some sample points on the diffuse reflection texture, sets the specular intensity parameter and roughness parameter for these points, and then calculates these two specular parameters on the basis of the diffuse similarity between the other positions on the diffuse reflection texture and the selected sample points. Normal map module uses shading map to estimate a normal map. First, the shape from shading technique based on linear approximation is used to estimate the relative height of the organ surface from the shading map. Second, the organ surface height is filtered by Gauss blur to obtain the low-frequency features on the organ surface height. We estimate two normal maps by low frequency height feature and original height feature respectively. We overlay the two normal maps by linear interpolation to obtain the final normal mapWe observed that the concave and convex areas on the organ surface formed shadows in the shading map and these could be used as environmental shadows. Therefore, the system uses an illumination map S to generate a simple AO map. First, the brightness of the shading map is enhanced, and then, Gauss blur is used to get the final AO map. To facilitate users to evaluate the quality of the final product, that is the appearance of the texture, the real-time visualization module loads the generated appearance of the texture for real-time rendering to provide the users the WYSIWYG three-dimensional visualization results. Experiments show that our appearance of texture matches the appearance parameters in the three-dimensional rendering formula and can be directly applied to the generation of realistic digital images of plants in three-dimensional visualization software. Compared to the current commercial texture production software, the texture style generated by this system is more comprehensive, more in line with the production needs of digital images of plants, and more automated and user-friendly. It meets the need for rapid and accurate rendering of the textured appearance of plant organs and provides a technical tool for the development of three-dimensional digital image resources of agricultural specimens.