An Improved Multiplexed Metropolis Light Transport Algorithm Based on Variance Filtering

Aiming at the problem of start-up bias in the multiplexed Metropolis light transport（MMLT） algorithm, an improved algorithm based on variance filtering is proposed to make MMLT quickly achieve a stable distribution from any distribution. Firstly, the high-contribution sampling points obtained in the presampling stage are stored. Secondly, the variance of each sample is judged. Finally, a certain number of samples with high contribution and low variance are taken as the seed samples into the formal sampling. The results of experiments conducted in different scenarios using different algorithms show that the algorithm produces less noisy results under direct illumination and mixed illumination with less time.