The Study Of Bidirectional Path Tracing Based Metropolis Global Illumination Method

Monte Carlo method is the core of global illumination. There are many global illumination algorithms derived from basic Monte Carlo path tracing. But these methods are only optimized for a fairly narrow class of input scenes. Because they don't know the different path distributions in different scenes when sample the paths from the light source to the eye. All these methods sample the paths randomly. They will work inefficient when the paths in the input scenes distribute highly non-uniformly.In this paper, we sample the paths in path space using Metropolis sampling method. We also provide a bidirectional path tracing based Metropolis light tracer in our paper. At first, we introduce the render equation and Monte Carlo method. And then, we introduce the basic Metropolis sampling method. To render the image, we generate a seed path by bidirectional path tracing,. From this seed path, we generate a sequence of paths by mutating current path using our mutation methods. Each mutation is accepted or rejected with calculated acceptance probability. We add a sample to the pixel which accepted path belongs to. The sequence of path is a Markov chain. As the Markov chain grows, our sample distribution converges to a stationary distribution proportionate to the image contribution. We can get the render result by scaling sample distribution. We introduce our two path mutation methods. The paper also includes the details of our Metropolis light tracer.