Authors:
Hongliang Yuan
1
;
Changwen Zheng
2
;
Quan Zheng
1
and
Yu Liu
1
Affiliations:
1
Chinese Academy of Sciences and University of Chinese Academy of Sciences, China
;
2
Chinese Academy of Sciences, China
Keyword(s):
Adaptive Rendering, Adaptive Order Selection, Monte Carlo Ray Tracing, Mean Squared Error.
Related
Ontology
Subjects/Areas/Topics:
Computer Vision, Visualization and Computer Graphics
;
Image-Based Rendering
;
Rendering
Abstract:
We propose a new adaptive sampling and reconstruction method based on a novel, adaptive order polynomial
fitting which can preserve various high-frequency features in generated images and meanwhile mitigate the
noise efficiently. Some auxiliary features have strong linear correlation with luminance intensity in the smooth
regions of the image, but the relationship does not hold in the high-frequency regions. In order to handle
these cases robustly, we approximate luminance intensity in the auxiliary feature space by constructing local
polynomial functions with order varying adaptively. Firstly, we sample the image space uniformly. Then
we decide the order of fitting with the least estimated mean squared error (MSE) for each pixel. Finally,
we distribute additional ray samples to areas with higher estimated MSE if sampling budget remains. We
demonstrate that our method makes significant improvement in terms of both numerical error and visual quality
compared with the state-of-the-art.