Efficient interactive multi-resolution rendering for translucent materials under area lights
Koa, Ming Di
Date of Issue2017
School of Computer Science and Engineering
Translucent materials exist all around us. They can be commonly found in materials such as marble, jade and fluids such as milk. Subsurface scattering describes one of the most complex yet beautiful effects of light in translucent materials. However, this light and media interaction is difficult to simulate. Although Monte Carlo approaches can simulate these effects, they are computationally expensive. Further complexities in simulation occur, when indirect illumination from translucent materials are considered, as they scatter light and illuminate other surfaces. In addition, when translucent materials are illuminated by area lights, fast techniques such as voxel based methods have difficulties in estimating direct visibility. In my research, I aim to develop efficient techniques for rendering homogeneous translucent materials under area light illumination at interactive speed. In the first research of this thesis, I present a real-time approximation technique for subsurface scattering rendering that uses two separate voxel structures for rendering single scattering and multiple scattering effects. In my algorithm, a ray marching approach is used to traverse rays from the light source through the voxel structure, the Enhanced Subsurface Light Propagation Volumes (ESLPV), while leaving its radiance compressed into spherical harmonics in each marching step. During rendering, we use importance sampling to select important samples along the refracted ray. Some propagation optimization routines are used to remove unnecessary computations in the propagation iterations of Light Propagation Volumes (LPV) based algorithms and also accelerate the rendering time while maintaining the same visual quality. As the LPV is not a physically accurate algorithm, flux can be lost unnecessarily for each simulation step. I proposed a multi-resolution voxel structure to reduce the flux lost due to the numerous scattering events simulated in high resolution voxels and also preserve the radiance for backlighting at similar quality to the Photon Based Diffusion methods. In the second research, I proposed a deferred multi-resolution approach for rendering direct illumination effects such as soft shadows when translucent and other opaque objects are illuminated by area lights. My approach subdivides the screenspace into multiresolution 2D-fragments in which higher resolution fragments are created to represent geometric and depth discontinuities as well as shadow boundaries. I proposed a sub-fragment visibility test scheme that focuses on discontinuities within a fragment. A gradient aware subdivision scheme is developed to refine the discontinuities according to its gradients. My technique utilizes the stream-compaction feature of the transform feedback shader (TFS) in the graphics shading pipeline to filter out fragments for soft shadow refinement easily. A single pass screenspace irradiance upsampling scheme which uses radial basis functions (RBF) and an adaptive variance scaling factor is propose for interpolating scattered fragments. This reduces artifacts caused by large fragments while also significantly reducing the number of visibility rays that we require. In my final research, I combine the work of my first two researches and tackle two separate illumination problems for area lighting. First, the focus is on the appearance of translucent materials under illumination from area lights, which also casts soft shadows on underlying surfaces. Next, the translucent materials distribute this scattered lighting towards its surrounding surfaces. Next, we focus on the appearance of indirect illumination on the surrounding scene after receiving scattered light from translucent materials. To solve both direct and indirect illumination problems, we use a set of Poisson disk distributed samples on translucent objects and scene to gather illumination from area lights. The gathered illuminations are ’injected’ as transmitted flux into the ESLPV to render the translucent objects. For simulating indirect illumination, the reflected flux are injected into the LPV and distributed. The multi-resolution approach employed in this thesis provides accelerated rendering which can be used in various multimedia applications (e.g. games that require translucent object rendering or soft shadow rendering), and visual arts (e.g. pre-renders for rendering programs).
DRNTU::Engineering::Computer science and engineering::Computing methodologies::Computer graphics