Adaptive cluster rendering via regression analysis

Monte Carlo ray tracing suffers noise and aliasing because of low sampling rate. We show that sparse samples can be used to generate high quality images based on feature cluster and regression analysis. Our algorithm has two main stages: adaptive sampling and polynomial reconstruction. In sampling stage, rendering space are organized into clusters based on their features. A feature vector is used to distinguish the different features, which contains gradient, variance and position. Clusters are progressively modified by adaptive sampling. In reconstruction stage, we model each cluster by smooth polynomial functions using regression analysis. The final image is synthesized by integrating these functions. The experiments show that our algorithm generates higher quality images than the previous methods.