Antialiasing for Automultiscopic Displays

1
Antialiasing for Automultiscopic Displays

• Matthias Zwicker, Wojciech Matusik, Fredo Durand,
  Hanspeter Pfister

2
Automultiscopic displays
2D display
Automultiscopic display
View independent pixel
View dependent pixel
3
Automultiscopic displays
2D display
Automultiscopic display
View independent pixel
View dependent pixel
4
Automultiscopic displays

• Stereoscopic viewing without glasses

5
Automultiscopic displays

• Stereoscopic viewing without glasses

6
Automultiscopic displays

• Parallax without head tracking
• Window to a virtual world

Automultiscopic display
7
Implementation

• Lenticular sheets, parallax barriers Ives 1931

High resolution display
Parallaxbarrier
Slit
Right eye
Left eye
8
Implementation

• Newsight display
• 640x384 view dependent pixels
• Horizontal parallax only
• 8 directions per pixel
• 30 degree viewing zone

Newsight
9
Aliasing

• Simulated views, horizontal translation of
  viewpoint
• Aliasing in the directional domain appears as
  temporal aliasing

10
Overview

• Previous work
• Antialiasing
• Mapping images to the display
• Image acquisition

Acquisition and rendering of image data
forautomultiscopic displays
11
Previous work

• Antialiasing for automultiscopic displays
• Based on wave optics or geometric considerations
• Correcting interperspective aliasing in
  auto-stereoscopic displays, Moller and Travis,
  TVCG 2005
• Holographic stereograms as discrete imaging
  systems, Halle, 1994
• No practical algorithms for high-quality
  rendering without depth information

12
Previous work
Rendering light fields on automultiscopic
displays

• Dynamically reparameterizable light-fields,
  Isaksen et al., SIGGRAPH 2000

Two plane parameterization
Display
Parallaxbarrier
Emitted rays
13
Previous work
Rendering light fields on automultiscopic
displays

• Dynamically reparameterizable light-fields,
  Isaksen et al., SIGGRAPH 2000
• No display antialiasing

Two plane parameterization
Display
Parallaxbarrier
14
Previous work
Signal processing analysis of light fields

• Plenoptic sampling, Chai et al., SIGGRAPH 2000
• Extension to include displays

15
Signal processing analysis of light fields
Ray
Two plane parameterization
Ray space
Chai et al., 2000
16
Signal processing analysis of light fields
Two plane parameterization
Ray space
Chai et al., 2000
17
Signal processing analysis of light fields
Scene with constant depth
Two plane parameterization
Ray space
Chai et al., 2000
18
Signal processing analysis of light fields
Fourier transform
Light field spectrum
Ray space
Chai et al., 2000
19
Signal processing analysis of light fields

• Scenes with bounded depth range

Light field spectrum
Chai et al., 2000
20
Antialiasing

• Signal processing approach Chai et al.
• Display bandwidth and depth of field
• Resampling

21
Display bandwidth
Two plane parameterization
Display
Parallaxbarrier
Emitted rays
22
Display bandwidth
Two plane parameterization
Display
Parallaxbarrier
Sampling grid in ray space
23
Display bandwidth
Display bandwidth,Nyquist limit
Frequency domain
Angular bandwidth
Spatial bandwidth
24
Display depth of field
Display bandwidth
Frequency domain
Display depth offield
Distance betweendisplay surface and parallax
barrier
25
Display depth of field
Display bandwidth
Frequency domain
Display depth offield
Outside display depth of field
26
Display depth of field

• Numerical example
• Display specifications
• LCD display, pixel pitch 0.25mm
• Spacing of slits 2mm
• Distance to parallax barrier 4mm
• Depth of field /- 32mm
• Shallow depth of field, but sufficient in practice

27
Resampling
Cameraarray
Display
Input coordinates
Display coordinates
28
Resampling
Cameraarray
Ray space(display coordinates)
Display
Camera samplesDisplay samples
29
Resampling

• Avoiding aliasing artifacts
• Heckberts framework, texture mapping Heckbert
  1989
• Reconstruction aliasing
• Pre-aliasing

30
Resampling
Input spectrum
Replicas
Input spectrum(camera coordinates)
31
Resampling
Input spectrum
Replicas
Reconstruction Stewart et al. 2003
32
Resampling
Input spectrum
Replicas
Displaybandwidth
Reconstruction
Reparameterization
33
Resampling
Input spectrum
Replicas
Displaybandwidth
Prefiltering,sampling
Reconstruction
Reparameterization
34
Resampling

• Implementation
• Spatial domain
• Combined resampling filter

35
Resampling results

• Simulated views, horizontal translation of
  viewpoint

No display prefiltering
Full resampling
36
Resampling results

• Simulated views, horizontal translation of
  viewpoint

No display prefiltering
Full resampling
37
Mapping input data to the display

• Naïve approach
• Uniform scaling of camera geometry
• Depth range of real scenes often exceeds display
  depth of field
• Prone to blurry results

38
Mapping input data to the display
Inputspectrum
Naïve mapping
Camera coordinates
Displaycoordinates
Display band-width, DOF
Min. depth
Max. depth
39
Mapping input data to the display
Inputspectrum
Naïve mapping
Camera coordinates
Displaycoordinates
Display band-width, DOF
Min. depth
Max. depth
40
Mapping input data to the display
Adapting todisplay DOF
Inputspectrum
Naïve mapping
Camera coordinates
Displaycoordinates
Displaycoordinates
Display band-width, DOF
Min. depth
Max. depth
41
Mapping input data to the display
Locomotive indisplay depth of field
Naïve mapping
42
Acquisition

• Avoid trial and error when setting up camera
  arrays
• Minimum acquisition sampling requirements
• Scene parameters
• Target display
• For optimal image quality, non-central replicas
  of input light field must not intersect display
  prefilter

43
Acquisition

• Tightest packing of replicas of the input light
  field

Undersampling
Inputspectrum
Displayprefilter
Usabledisplaybandwidth
Display coordinates
44
Acquisition

• Tightest packing of replicas of the input light
  field

Minimum sampling
Inputspectrum
Displayprefilter
Usabledisplaybandwidth
Display coordinates
45
Acquisition

• Java Applet graphics.ucsd.edu/matthias/Supplement
  alMaterial/AntialiasingFor3DDisplays/

Visualization
46
Acquisition

• Java Applet graphics.ucsd.edu/matthias/Supplement
  alMaterial/AntialiasingFor3DDisplays/

Visualization
47
Conclusions
Acquisition and rendering of image data
forautomultiscopic displays

• Signal processing analysis of automulti-scopic
  displays
• Display bandwidth and depth of field
• Antialiasing
• Adapting display depth of field
• Acquisition guidelines

48
Future work

• User study to evaluate effectiveness of
  antialiasing scheme
• Animated scenes
• Determine perceptionally optimal filter
• Multiview video compression

49
Questions
50
(No Transcript)
51
Mapping input data to the display
Camera array
Display
Acquiredgeometry
52
Mapping input data to the display
Camera array
Display
Acquiredgeometry
Observedgeometry
53
Mapping input data to the display
Camera array
Display
Acquiredgeometry
Observedgeometry

• Compute and such that given depth range is
  in depth of field of the display