RENDERING OF INHOMOGENEOUS VOLUMES USING PERTURBATION FUNCTIONS

Abstract

Modeling of light transmission in heterogeneous volumes is of great importance in many fields, such as medical imaging, scientific visualization and synthesis of realistic images. Visual effects use complex three-dimensional structures such as smoke and clouds. However, modeling light transmission requires many calculations. For example, Monte-Carlo methods, which are based on path tracing, require the construction of a huge number of light paths. At the same time, each light path consists of thousands of scattering parts. A method for rendering inhomogeneous volumes using perturbation functions is presented. An approach is proposed for sampling light transmission paths in inhomogeneous media. The approach is based on the radiation transfer equation, using the integral formulation of the direct scattering algorithm. Bounding shells based on perturbation functions are used. To speed up calculations an inhomogeneous medium is divided into homogeneous and residual parts. The residual part is the difference between an inhomogeneous and homogeneous medium. For a homogeneous part light transmission paths are constructed in an analytical form. Next, the path-tracing algorithm is used. Samples in the light transmission path in the homogeneous and residual parts are made separately. This minimizes the costly calculations of direct scattering coefficients that change when traversing space. The method has advantages in comparison with approaches using an octal tree, with a large volume resolution the efficiency of calculations increases. The results of the work are integrated into the path tracer. Objects based on perturbation functions as an acceleration structure are used. The empty space is determined and approximate local extremes of the base volumes are stored. Objects based on perturbation functions adapt to volume uniformity. Voluminous data sets based on voxels are stored. Performance is compared using the number of queries, visualization time, root mean square error and metrics, that is, the search in units of variance.