Research On High-fidelity Rendering Method Based On Metropolis Light Transport

Metropolis light transport algorithm is a kind of unbiased Markov Chain Monte Carlo Global illumination algorithm,which has high path search efficiency in some complex lighting scenes and rendering effect is good and robust.However,in the scene where the local light-shade mixed,the algorithm has the problem of asymmetry of the number of samples and the illuminations distribution.In addition,the algorithm ignores the influence of the noise level in the pixel on the rendering result during the sampling process.These features will result in insufficient fidelity of the local detail position of the final rendered image.Therefore,three improvements schemas have been proposed progressively.The first constructs a sampling weight on the sample quantity and the illumination from a relatively macroscopic angle,taking into account the influence of the sample quantity and the illumination distribution,such optimizing the symmetry of the sample quantity and the illumination distribution.The second is to further accurately measure the state of the samples in the pixel from the microscopic angle of pixel synthesis,and then propose an adaptive-sampling algorithm based on the ambiguity.During the sampling process,the samples in each pixel are used as a set,and the ambiguity is used to evaluate the noise-level within the pixel,and then use the ambiguity as the sampling weight for the different noise-level positions.The third is to further optimize the initial samples of the pre-sampling phase of the second scheme,so the method of variance filtering is introduced to optimize the sample seeds that enter the formal rendering phase.In addition,in the formal rendering phase,the ambiguity remains the sampling weight for different noise-level positions.The experimental results show that the first improvement scheme improves the local symmetry of the number of samples and the illumination distribution in the scenes with different illumination and material where light-dark mixed,and improves the rendering fidelity;The second scheme introduces the fuzzy degree as the weight of the adaptive sampling algorithm,which can improve the rendering noise while optimizing the overall sampling distribution,and achieve the effect of adaptive sampling during the post-rendering;The third scheme combines the variance filtering on the basis of the ambiguity,and has certain optimization effects on the noise of the rendering result.At last the thesis is summarized and the follow-up research work is put forward depending on the analysis for some questions.