Study On Image Denoising Methods Based On Metropolis Light Transport

Metropolis light transport is an unbiased and robust Monte Carlo method, which could efficiently reduce random noise during rendering realistic graphics under global illumination. Once the efficient paths which contribute to the final image have been found, the algorithm will continue sampling the adjacent area to make the sampled paths have greater influence to the final results. Where, Metropolis sampling is based on a certain probability distribution determined by some importance functions, and the extension for path space sampling is regional and which does small scale change for the current path by mutation. So the extension made by importance function and the reasonable selection of mutation strategy could effectively improve the rendering results of Metropolis light transport.Based on the above two important factors which influence Metropolis light transport algorithm, this paper proposes two improved algorithms. One is the improved Metropolis light transport based on multiple importance sampling. It mostly combine multiple importance sampling to improve Metropolis light transport to better solve the problems about the large correlation and high variance among samples caused by choosing improper importance functions better. The experiences illustrate that the quality of images generated by the improved algorithm is better compared with Metropolis light transport in the same scenes background. The other is the improved Metropolis light transport based on mutation strategy. In order to avoid making the path sampling get stuck into region and make the sampled paths of path space satisfy ergodicity, we improve the path mutation strategy that whether to continue mutating at the adjacent area of the current path or to randomly generate new path sample as the initial path sample by the contribution of the current path to the final image. The experiments results show that the improved algorithm can traverse path space better and generate the image with less noise comparing to Metropolis light transport algorithm and PSSMLT(Primary Sample Space Metropolis Light Transport) algorithm.Finally, we summarize this paper and make an outlook of research for the next step.