Adaptive Rendering Using a Best Matching Patch

ISCAS OpenIR

	Adaptive Rendering Using a Best Matching Patch
	Liu, Y; Zheng, CW
	2016
发表期刊	IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS
ISSN	1745-1361
卷号	E99D 期号:7 页码:1910-1919
摘要	A novel rendering algorithm with a best-matching patch is proposed to address the noise artifacts associated with Monte Carlo renderings. First, in the sampling stage, the representative patch is selected through a modified patch shift procedure, which gathers homogeneous pixels together to stay clear of the edges. Second, each pixel is filtered over a discrete set of filters, where the range kernel is computed using the selected patches. The difference between the selected patch and the filtered value is used as the pixel error, and the single filter that returns the smallest estimated error is chosen. In the reconstruction stage, pixel colors are combined with features of depth, normal and texture to form a cross bilateral filter, which highly preserves scene details while effectively removing noise. Finally, a heuristic metric is calculated to allocate additional samples in difficult regions. Compared with state-of-the art methods, the proposed algorithm performs better both in visual image quality and numerical error.; A novel rendering algorithm with a best-matching patch is proposed to address the noise artifacts associated with Monte Carlo renderings. First, in the sampling stage, the representative patch is selected through a modified patch shift procedure, which gathers homogeneous pixels together to stay clear of the edges. Second, each pixel is filtered over a discrete set of filters, where the range kernel is computed using the selected patches. The difference between the selected patch and the filtered value is used as the pixel error, and the single filter that returns the smallest estimated error is chosen. In the reconstruction stage, pixel colors are combined with features of depth, normal and texture to form a cross bilateral filter, which highly preserves scene details while effectively removing noise. Finally, a heuristic metric is calculated to allocate additional samples in difficult regions. Compared with state-of-the art methods, the proposed algorithm performs better both in visual image quality and numerical error.
收录类别	SCI
关键词	Adaptive Sampling And Reconstruction Rendering Patch Shift Cross Bilateral Filter
部门归属	Chinese Acad Sci, Inst Software, Natl Key Lab Integrated Informat Syst Technol, Beijing 100864, Peoples R China.
语种	英语
WOS记录号	WOS:000381562700018
引用统计
内容类型	期刊论文
URI标识	http://ir.iscas.ac.cn/handle/311060/17322
专题	中国科学院软件研究所
推荐引用方式 GB/T 7714	Liu, Y,Zheng, CW. Adaptive Rendering Using a Best Matching Patch[J]. IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS,2016,E99D(7):1910-1919.
APA	Liu, Y,&Zheng, CW.(2016).Adaptive Rendering Using a Best Matching Patch.IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS,E99D(7),1910-1919.
MLA	Liu, Y,et al."Adaptive Rendering Using a Best Matching Patch".IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS E99D.7(2016):1910-1919.