Authors:
Changwen Zheng
1
and
Yu Liu
2
Affiliations:
1
Chinese Academy of Sciences, China
;
2
Chinese Academy of Sciences and University of Chinese Academy of Sciences, China
Keyword(s):
Adaptive Rendering, Compressed Sensing, Ray Tracing, Cross-bilateral Filter.
Related
Ontology
Subjects/Areas/Topics:
Computer Vision, Visualization and Computer Graphics
;
Geometry and Modeling
;
Physics-Based Modeling
;
Rendering
;
Rendering Algorithms
Abstract:
Monte Carlo renderings suffer noise artifacts at low sampling rates. In this paper, a novel rendering algorithm
that combines compressed sensing (CS) and feature buffers is proposed to remove the noise. First, in the
sampling stage, the image is divided into patches that each one corresponds to a fixed resolution. Second, each
pixel value in the patch is reconstructed by calculating the related coefficients in a transform domain, which is
achieved by a CS-based algorithm. Then in the reconstruction stage, each pixel is filtered over a set of filters
that use a combination of colors and features. The difference between the reconstructed value and the filtered
value is used as the estimated reconstruction error. Finally, a weighted average of two filters that return the
smallest error is computed to minimize output error. The experimental results show that the new algorithm
outperforms previous methods both in visual image quality and numerical error.