Realtime Rendering for Adjustable Plenoptic Illumination Functions

Description

Plenoptic illumination function (PIF) or bidirectional reflectance distribution function (BRDF) allows us to light a scene under complex illumination conditions. However, the brute-force way to integrate all incoming radiances is time-consuming. To address this issue, the researchers can use basis function techniques, such as spherical harmonic (SH) or cubemap based spherical wavelet (CSW) approaches, to increase the rendering speed.The SH approach can only efficiently handle low frequency lighting effects and rotating environment. Although the CSW can efficiently handle all-frequency lighting effects, it suffers from the uneven sampling problem and aliasing-error. Lastly, existing basis function techniques can only provide real-time rendering for either fixed BRDF with time-varying environment or editable BRDF with a fixed environment.This project has two phases. In phase one, the researchers will develop a computational-friendly uniform spherical wavelet (SW) approach for time-varying illumination. They thus can efficiently handle local high frequency lighting features and reduce the number of wavelet coefficients needed. To accelerate the process and to reduce the data size held in the graphics hardware, rendering-friendly data compression methods for wavelet coefficients will be developed.In phase two, the researchers will investigate a novel vector graphics representation for the visibility. With the use of summed area table for representing the lighting environment, the proposing approach can efficiently and effectively handle real-time rendering for view-dependent BRDF or PIF. Analogous to dimensionless property of vector graphics, they can raise the effective sampling rate of visibility by a few orders at a very low cost. To further improve the quality, the researchers will also develop a morphing method to handle visibility functions among adjacent vertices. Hence, they have a high quality rendering result, even the 3D object is not densely defined. By combining the result of phase one, the researchers expect that the proposing approach can handle a wide range of BRDF models, including BRDF data captured from real object. Since the lighting environment, visibility and BRDF are individually represented by different techniques, real-time rendering for editable BRDF with time-varying environment becomes possible.