Research On Metropolis Global Illumination Rendering Method

Metropolis light tracing algorithm is an unbiased global illumination algorithm.It is based on the Markov chain Monte Carlo method.It shows strong processing power in complex scenes with only a small amount of effective light or a large number of indirect lighting,so it has attracted widespread attention in the photorealistic graphics rendering methods.However,due to the correlation between the samples during the sampling process of the Metropolis light transport algorithm,the time it takes for the Markov chain to converge to a steady-state distribution varies and cannot be estimated,resulting in pixel differences affecting the rendering results.At the same time,when the algorithm is rich in local structural information,there will be a problem that the sampling relies too much on radiance and ignores the structural information.Based on the above problems,this paper proposes three improvement methods.The first method is an improved multiplexed Metropolis light transport algorithm based on grouping.This method introduces group sampling technology to modify the composition of the Markov chain in the algorithm.It uses a global sampler to concentrate the sampling distribution information,thereby transforming the acceptance probability function from sample-to-sample to chain-to-chain.Finally,the convergence of a single sample becomes the convergence of the entire chain,so that the sampling distribution gradually approaches the objective function,the convergence time is fixed to a certain interval,and the pixel difference is reduced.The experimental results show that the improved algorithm can reduce the pixel variance under the same conditions and make the overall effect of the rendered image more realistic.The second method is an improved multiplexed Metropolis light transport algorithm based on information entropy.This method first introduces the information entropy capable of expressing the structure into the importance function to balance the influence of radiance,so as to improve the sampling weight of locations with rich structural information but low radiance.Secondly,the mutation probability is dynamically modified during the mutation process,so that limited sampling resources are better optimized.The experimental results show that the improved algorithm can optimize the profile information,and achieve the purpose of adaptive sampling during rendering in the scene under the same conditions.The third method is to merge the two improved algorithms.By modifying the Markov chain composition and re-weighting the importance function,the two methods work together to reduce the variance and optimize the overall image structure.Experimental results show that the improved fusion algorithm makes the rendered image brighter and reduces the blurriness.Finally,the full text is summarized,and the follow-up research work is expected.