View
220
Download
0
Category
Tags:
Preview:
Citation preview
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
A study on resolution-improved volumetric computational reconstruction using stereo
matching in 3D integral imaging
Graduate school of Design & IT Dongseo Univ
Hwang Yong-Hyeonhttp://kowon.dongseo.ac.kr/~d0066042
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Contents
1. Introduction2. Conventional method3. Proposed method4. Experiment Results5. Future work
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Introduction
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
3D Display
Human Visual System
Stereo Disparity
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
3D Display
1.Stereoscopic display
2.Multi-view display
3.Integral imaging display
4.Holographic display
Leftcamera
Rightcamera
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
3D Display
1.Stereoscopic display
2.Multi-view display
3.Integral imaging display
4.Holographic display
<Cambridge Univ.>
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
3D Display
1.Stereoscopic display
2.Multi-view display
3.Integral imaging display
4.Holographic display
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
3D Display
1.Stereoscopic display
2.Multi-view display
3.Integral imaging display
4.Holographic display
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Integral imaging
Pickup Display3D 3D
Elemental images
Lenslet array
3D object 3D image
Elemental images
Lenslet array
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Optical vs Computational
Optical pickup
OpticalReconstruction
ComputationalReconstruction
Real 3D display(NHK,SNU, Connecticut Univ. etc)
Pattern recognition(Connecticut Univ., KWU etc)
VCR
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
CII system
Pickup process(optical)
Volumetric computational reconstruction (VCR)(computer)
plane by plane
VCR
<Computational integral imaging(CII) system>
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
VCR of 3D objects
Elemental image Reconstruction Object Pinhole array
Elemental image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Conventional method
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Conventional method
B. Javidi et al, R. Ponce-Diaz, and S.-H. Hong, Opt. Lett. 31, 1106-1108 (2006)
Reference plane image
z=zrz=0
Cross-correlation
z=zrz=0
Target plane image
Pickup device
Reference CII system
zr
Lenslet array3D object
Pickup device
Target CII
system
zr
Lenslet array
zo1
3D object occlusion
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Proposed method
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Novel Proposed method
Sub-Imagetransformation
Segmentation using stereo matching
VCR
zr
Inverse Sub-Image
transformation
Pickup device
Lenslet array
3D objectOcclusi
on
zrzo
Elemental image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Elemental image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Novel Proposed System
Sub-Imagetransforma
tion
Segmentation using stereo matching
VCR
zr
Pickup device
Lenslet array
3D objectOcclusi
on
zrzo
Inverse Sub-Image
transformation
Elemental image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Sub-image transformation
Elemental image Sub-image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Sub-image transformation
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Sub-image transformation
1 pixel sub-image transformation
Elemental imagesElemental images
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Sub-image transformation
900 x 900 1 pixel sub-image transformation
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Novel Proposed System
Segmentation using stereo matching
VCR
zr
Pickup device
Lenslet array
3D objectOcclusi
on
zrzo
Inverse Sub-Image
transformation
Elemental image
Sub-Imagetransforma
tion
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Stereo matching
PL = (X,Y,Z)OL
x
y
z (uL,vL)
OR
x
y
z
(uR,vR)
base
line B
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
feature-based method
Original image Feature image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Matching window
area-based method
Left image Right image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Stereo matching
SSD(Sum of square distance)
SAD(Sum of absolute distance)
NCC(Normalized correlation
coefficient)
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Disparity map
Right ImageLeft ImageDisparity
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Sub-image transformation
900 x900 5 pixel sub-image transformation
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Stereo matching
5 X 5 sub-image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Stereo matching
Left image Right image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Disparity map
Disparity map
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Segmentation result
5 x 5 Sub-image Segmentation result
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Novel Proposed System
Segmentation using stereo matching
VCR
zr
Pickup device
Lenslet array
3D objectOcclusi
on
zrzo
Inverse Sub-Image
transformation
Elemental image
Sub-Imagetransforma
tion
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Inverse sub-image
5 x 5 Sub-image
Elemental image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Novel Proposed System
Segmentation using stereo matching
VCR
zr
Pickup device
Lenslet array
3D objectOcclusi
on
zrzo
Inverse Sub-Image
transformation
Elemental image
Sub-Imagetransforma
tion
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
VCR
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Experiment Results
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
VCR of conventional method
Elemental image VCR image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
VCR of proposed method
Segmentation elemental image VCR image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
PSNR of conventional method
PSNR : 20.94db
Reference image Reconstruction image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
PSNR of proposed method
PSNR : 29.61db
Reference image Reconstruction image
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Test image
Cow Toy Car Horse
Cow Toy Car Horse
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
PSNR
Conventional method
Proposed methodImproved
PSNR
Cow 20.94 29.61 8.67db
Toy 22.31 29.81 7.5db
Car 18.03 24.01 5.98db
Horse 13.78 23.19 9.41db
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Experiment environment
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Real image
Elemental image
Sub-image a die Toy
Segmentation result
Hwang Yong-Hyeon Dongseo Univ. E-mail:pial051@dit.dongseo.ac.kr
Future work
• Real image test• Segmentation using difference stereo
matching• Real image calibration
Recommended