Comparing ambient occlusion sampling methods in a production level offline rendering application
NURMI, RAMI (2008)
Tässä tietueessa ei ole kokotekstiä saatavilla Treposta, ainoastaan metadata.
NURMI, RAMI
2008
Tietojenkäsittelyoppi - Computer Science
Informaatiotieteiden tiedekunta - Faculty of Information Sciences
Hyväksymispäivämäärä
2008-07-17Tiivistelmä
Global illumination is a general term for various methods used in 3D graphics to simulate light coming directly and indirectly to a surface location. The computational cost of the full-scale global illumination is very high and rendering times for the scenes using global illumination can range from minutes to hours per image. The high computational cost has led to the development of many approximations. Ambient occlusion is a technique that approximates the indirect global illumination at significantly lower cost. The lower computational cost has made ambient occlusion a popular technique used by movie industry and more recently it has gathered the interest of real-time graphics industry. In this study, five sampling methods that can be used in ray traced ambient occlusion computations are reviewed and compared: random, jittered stratified, Poisson disk, Halton low discrepancy sequence and Hammersley low discrepancy sequence sampling. The results of this study indicate that Hammersley sequence p1=2 is the second slowest but gives the best results in terms of quality. Halton sequence p1=2, p1=7 sampling and jittered stratified sampling perform almost equally in terms of image quality, but the Halton sequence is notably slower of the two. This study also represents a simple Halton sequence based optimization to ray traced ambient occlusion computation. The optimisation gives a notable improvement in the rendering speed and has no adverse effects to the image quality. The results indicate that optimised Halton sequence p1=2, p1=7 gives the second best image quality and is the fastest of the sampling methods tested. The optimisation is very easy to implement and has a low memory cost which makes it an attracting method of speeding up ambient occlusion computations.