High quality depth map estimation of object surface from light-field images

doi:10.1016/j.neucom.2016.09.136

CASIA OpenIR > 模式识别实验室

	High quality depth map estimation of object surface from light-field images
	Liu, Fei1,2 ; Hou, Guangqi2 ; Sun, Zhenan2 ; Tan, Tieniu2 ; Guangqi Hou
发表期刊	NEUROCOMPUTING
	2017-08-23
卷号	252 期号:252 页码:3-16
文章类型	Article
摘要	; Light-field imaging provides a novel solution to the passive 3D imaging technology. However the dense multi-view sub-aperture images decoded from the light-field raw image have extremely narrow baselines, which lead to inconsistent matching with terrible blurriness and ambiguities. This paper presents an accurate depth estimation algorithm for object surface using a lenslet light-field camera. The input data for depth estimation can be both light-field videos and images under indoor and outdoor environment. To tackle the continuously changing outdoor illumination and take full advantage of rays, rendering enhancement is performed through denoising and local vignetting correction for obtaining high-fidelity 4D light fields. The novel sub-aperture image pair selection and stereo matching algorithm are proposed for disparity computation. Then we apply the disparity refinement for recovering high quality surface details and handling disparity discontinuities. Finally both commercial and self-developed light-field cameras are used to capture real-world scenes with various lighting conditions and poses. The accuracy and robustness of the proposed algorithm are evaluated both on synthetic light-field datasets and real-world scenes by comparing with state-of-the-art algorithms. The experimental results show that high quality depth maps are recovered with smooth surfaces and accurate geometry structures. (C) 2017 Elsevier B.V. All rights reserved.
关键词	Light Field Depth Estimation Stereo Matching Disparity Refinement
WOS标题词	Science & Technology ; Technology
DOI	10.1016/j.neucom.2016.09.136
关键词[WOS]	BELIEF PROPAGATION ; STEREO
收录类别	SCI
语种	英语
项目资助者	National Basic Research Program of China(2012CB316300) ; National Natural Science Foundation of China(61420106015 ; 61302184 ; 61273272)
WOS研究方向	Computer Science
WOS类目	Computer Science, Artificial Intelligence
WOS记录号	WOS:000405884500002
引用统计	被引频次：19[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://ir.ia.ac.cn/handle/173211/19632
专题	模式识别实验室
通讯作者	Guangqi Hou
作者单位	1.Univ Chinese Acad Sci, Sch Engn Sci, Beijing, Peoples R China 2.Chinese Acad Sci, CAS Ctr Excellence Brain Sci & Intelligence Techn, Ctr Res Intelligent Percept & Comp, Natl Lab Pattern Recognit,Inst Automat, Beijing, Peoples R China
第一作者单位	模式识别国家重点实验室
推荐引用方式 GB/T 7714	Liu, Fei,Hou, Guangqi,Sun, Zhenan,et al. High quality depth map estimation of object surface from light-field images[J]. NEUROCOMPUTING,2017,252(252):3-16.
APA	Liu, Fei,Hou, Guangqi,Sun, Zhenan,Tan, Tieniu,&Guangqi Hou.(2017).High quality depth map estimation of object surface from light-field images.NEUROCOMPUTING,252(252),3-16.
MLA	Liu, Fei,et al."High quality depth map estimation of object surface from light-field images".NEUROCOMPUTING 252.252(2017):3-16.

条目包含的文件		下载所有文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
NeuroComputing2017.p（6100KB）	期刊论文	作者接受稿	开放获取	CC BY-NC-SA	浏览下载