Introduc)on to light field processing
Laboratoire de Communica0ons Audio-Visuelles (LCAV) Ecole Polytechnique Fédérale de Lausanne (EPFL)
Loïc Baboulaz 2015
“The most perfect photograph currently shows only one aspect of reality; it reduces to a unique image fixed on a plane, as a drawing or a pain0ng would be traced by hand”
Jonas Ferdinand Gabriel Lippmann (16 August 1845 – 13 July 1921)
Jonas Ferdinand Gabriel Lippmann (16 August 1845 – 13 July 1921)
• Part 1: Light, op/cs & light-field concepts • Part 2: Tradi/onal camera technology • Part 3: Light field camera technology
Jonas Ferdinand Gabriel Lippmann (16 August 1845 – 13 July 1921)
“Can we ask Photography to render all the richness that the direct view of an object offers?”
• Part 4: Sampling & processing the light field
Light
• Part 1: Light, op/cs & light-field concepts
Nature of light
ElectromagneGc radiaGon (wave vs. parGcle)
Main light properGes
Intensity PropagaGon direcGon Wavelength or color PolarizaGon
Light propagaGon theories
Quantum theory (wave packets) Wave theory (Maxwell’s equaGons) Ray-tracing (geometric opGcs)
• Part 2: Tradi/onal camera technology • Part 3: Light field camera technology • Part 4: Sampling & processing the light field
Ray-tracing
Light Phenomena
Principle
Complicated nature of light is reduced to a large set of narrow beams, or rays. Approximate soluGon of Maxwell’s equaGons
AssumpGon
Objects much bigger than wavelength of light
Pros
ComputaGonally efficient Geometric opGcs Well-suited for opGcal design
Cons
Cannot model diffracGon
ReflecGon RefracGon
DiffracGon
Light field
Light field parametric representa/on
• For every point in space, the light field L(r) describes the amount of light that is travelling at any instant of 0me
• For every point in space, the light field L(r) describes the amount of light that is travelling at any instant of 0me • A 2D light ray can be parameterized as: - Two-line intersec0ons: - Line intersec0on/slope:
, or as:
L(x, u) = L(x, x + d ⋅ p)
Light field parametric representa/on
Light field parametric representa/on
• For every point in space, the light field L(r) describes the amount of light that is travelling at any instant of 0me • A 2D light ray can be parameterized as:
• For every point in space, the light field L(r) describes the amount of light that is travelling at any instant of 0me • A 2D light ray can be parameterized as: , or as:
- Two-line intersec0ons: - Line intersec0on/slope:
, or as:
• A 3D light ray is described by 4 dimensions: , or as:
• Conversion between 2 representa0ons is straighYorward:
L(x, u) = L(x, x + d ⋅ p)
Light field parametric representa/on
Light field parametric representa/on
• Cartesian ray-space diagram is used to visualize rays, beams and complete light field:
• Cartesian ray-space diagram is used to visualize rays, beams and complete light field:
?
Light field parametric representa/on
Light field parametric representa/on
• Cartesian ray-space diagram is used to visualize rays, beams and complete light field:
• Cartesian ray-space diagram is used to visualize rays, beams and complete light field:
?
Light field parametric representa/on
Light field parametric representa/on
• Cartesian ray-space diagram is used to visualize rays, beams and complete light field:
• Cartesian ray-space diagram is used to visualize rays, beams and complete light field:
“Light beams share same directions at plane x ”
“Light beams overlap at plane u ”
Simple op/cal components
Simple op/cal components
• Propaga0on at flat interface:
• Propaga0on at flat interface:
• Converging lens vs. diverging lens optical axis
Simple op/cal components
optical axis
Lens parameters
• Propaga0on at flat interface: • Converging lens vs. diverging lens
Front focal point
optical axis
optical axis
• Naming conven0on for lenses: optical axis
biconvex
biconcave
positive meniscus
Back
negative meniscus
planoconvex
planoconcave
optical axis
Lens parameters
Thin Lens approxima/on
Front
Back focal point
Front
optical axis
• Lensmaker’s equa0on:
Back focal point
optical axis
• Thin lens equa0on:
• Convexity/concavity ó Sign of focal length ó Sign of lens radii. • P: lens power in dioptres; f: focal length in meters.
Thin Lens approxima/on
Thin Lens imaging Front
Front
Back focal point
• Thin lens equa0on:
optical axis
Back
optical axis
• Thin lens formula:
Example of propaga/on in thin lens
Example of propaga/on in thin lens
?
Example of propaga/on in thin lens
?
Example of propaga/on in thin lens
“Shearing of the light field”
Example of propaga/on in thin lens
Example of propaga/on in free space T
? “Shearing of the light field”
“Ray transfer matrix”
Example of propaga/on in free space
Example of propaga/on in free space T
T
?
Example of propaga/on in free space
Example of propaga/on in free space T
T
“Shearing of light field”
“Shearing of light field”
Ray transfer matrix
Ray transfer matrix
• Ray transfer matrices can be used to propagate light field forward and backward
• Ray transfer matrices can be used to propagate light field forward and backward
Propaga0on in free space with no scabering
Propaga0on in free space with no scabering
Refrac0on at flat interface
Refrac0on at flat interface
Reflec0on at flat mirror
Reflec0on at flat mirror
Iden0ty matrix
Refrac0on at thin lens
Refrac0on at thin lens
S h e a r matrix
Refrac0on at thick lens
Refrac0on at thick lens
S h e a r matrix Scaling matrix
Ray transfer matrix: example with concatenated op/cal components
Front
Ray transfer matrix: example with concatenated op/cal components
Back
Front
optical axis
Ray transfer matrices
Back
optical axis
Ray transfer matrices
Operations on light field
Operations on light field
Ray transfer matrix: example with concatenated op/cal components
Ray transfer matrix: example with concatenated op/cal components
Front
Back
Front
optical axis
Ray transfer matrices
Operations on light field
Back
optical axis
Ray transfer matrices
Operations on light field
Ray transfer matrix: example with concatenated op/cal components
Front
Ray transfer matrix: example with concatenated op/cal components
Back
Front
Back
optical axis
Ray transfer matrices
optical axis
Ray transfer matrices
Operations on light field
Operations on light field
Surface reflectances
Surface reflectances • The BRDF describes how light is reflected. It is material-specific. Source
Viewer
Light source
θs
θr
ϕr Scene Observer 1
Observer 2
ϕs
Surface reflectances
Surface reflectance: illustra/on
• The BRDF describes how light is reflected. It is material-specific. Source
Viewer
Rendering equa0on: Source
θs
θr
Viewer
θs
ϕr
ϕr
ϕs
ϕs
Surface reflectance: illustra/on
Source
Surface reflectance: illustra/on
Viewer
Source -π/2
θs
θr
0
+π/2
Diffuse (Lamber/an)
θr
Viewer -π/2
θs
ϕs
+π/2
Diffuse (Lamber/an)
θr
ϕr
0
ϕr ϕs
-π/2
0
+π/2
Diffuse + specular
Camera technology: the pinhole camera
• Part 1: Light, op/cs & light-field concepts • Part 2: Tradi/onal camera technology Anthony Browell, Rosy at Elkington Park 475
• Part 3: Light field camera technology • Part 4: Sampling & processing the light field
Pinhole camera
Pinhole camera
Image sensor
Aperture
Scene OpGcal axis
Scene
Image sensor O
OpGcal axis
Pinhole camera
Pinhole camera
Image sensor
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
x
OpGcal axis
x
u
• Ray-space parameterizaGon:
u
u
• Ray-space parameterizaGon: x
u x
Pinhole camera
Pinhole camera
Image sensor
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
x
OpGcal axis
u
• Ray-space parameterizaGon:
Depth u x
• Each light ray gives one image point
• Change of depth in the scene
Pinhole camera
Image sensor
Pinhole camera
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
OpGcal axis
Depth
Depth
• Change of depth in the scene
• Change of depth in the scene • All-in-focus image (~ infinite depth of field)
Pinhole camera
Pinhole camera
Image sensor
Aperture
Image sensor
Scene
I(x,y)
Aperture
Scene
I(x,y) OpGcal axis
OpGcal axis
P(X,Y,Z) f - “focal length”
• Change of depth in the scene • All-in-focus image (~ infinite depth of field) • Mapping from 3D to 2D:
Depth
P(X,Y,Z) f - “focal length”
Depth
• Change of depth in the scene • All-in-focus image (~ infinite depth of field) • Mapping from 3D to 2D:
Problem: the amount of light hiNng the sensor is very low, i.e. long exposure /me
Towards single lens camera
Image sensor
Towards single lens camera
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
OpGcal axis
“Open the aperture To allow more light in”
“Open the aperture To allow more light in”
Towards single lens camera
Image sensor
Towards single lens camera
Aperture
Image sensor
Scene
Aperture
OpGcal axis
Problem: blurred image points
Scene OpGcal axis
Problem: blurred image points à Solu/on: add collima/ng lens
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
OpGcal axis
Lens
Lens
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
OpGcal axis
sharp
Lens
sharp
• Each image point receives more light à lower exposure Gme • Image sensor integrates the incoming light cone
Scene OpGcal axis
Lens
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
OpGcal axis
Lens
blurry
Lens
sharp
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
OpGcal axis
blurry
Lens Focal plane
sharp
à Focal plane: points in object space that are focused by lens on image sensor
Scene OpGcal axis
blurry
Lens
sharp
How to change the focal plane?
Focal plane
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
sharp
blurry
blurry
Lens
OpGcal axis
sharp Focal plane
sharp
blurry
How to change the focal plane? By translaGng the posiGon of the image sensor
blurry
Lens Focal plane
sharp
How to change the focal plane? By translaGng the posiGon of the image sensor In other words: • When refocusing, the light field propagates differently in the camera before acquisiGon, • The image sensor captures the same light field, sheared differently for each focal plane.
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
sharp
OpGcal axis
Lens Focal plane
blurry
How to change the focal plane? By changing the lens configuraGon, i.e. refocusing.
CoC
blurry
Lens Focal plane
sharp
à Circle of Confusion (CoC): amount of acceptable blur for a point to be focused . CoC
CoC
Out-of-focus Focused Out-of-focus
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
Scene
OpGcal axis
OpGcal axis
Depth-of-field CoC
blurry
Depth-of-field
Lens
Lens Focal plane
Focal plane
sharp
à Depth-of-field: part of the scene that is focused well enough on image sensor. CoC
How to change the depth-of-field?
CoC
Out-of-focus Focused Out-of-focus
Single lens camera
Single lens camera
Image sensor
Aperture
Image sensor
Scene
Aperture
OpGcal axis
Scene OpGcal axis
Depth-of-field sharp
Lens Focal plane
CoC
How to change the depth-of-field? By closing the aperture of the lens.
Depth-of-field sharp
Lens
CoC
• Closing the aperture: 1. Extends the depth of field, 2. Reduces the amount of light, 3. Increases the exposure Gme.
Focal plane
Single lens camera: ray-space analysis
Image sensor
Single lens camera: ray-space analysis
Aperture
Scene Main lens
Image sensor
Aperture
Scene Main lens
OpGcal axis
OpGcal axis
x
F
u
Focal plane
Single lens camera: ray-space analysis
Image sensor
Focal plane
Single lens camera: ray-space analysis
Aperture
Scene Main lens
Image sensor
Aperture
OpGcal axis
x
F
u
u x
Ray-space diagram
OpGcal axis
x Focal plane
Scene Main lens
F
u Focal plane
Single lens camera: ray-space analysis
Image sensor
Single lens camera: ray-space analysis
Aperture
Scene
Image sensor
Main lens
Aperture
OpGcal axis
x
F
u
u x
Main lens OpGcal axis
x Focal plane
Scene
F
u Focal plane
u x
Ray-space diagram
Ray-space diagram
Single lens camera: ray-space analysis
Image sensor
Integral Projec/on Operator
Aperture
• Let be the canonical projec0on operator that reduces an N-dimensional func0on down to M-dimensions by integra0ng the last N-M dimensions [Ng06]:
Scene Main lens OpGcal axis
x
F
u Focal plane
u x
x Integral-projection
Ray-space diagram
Image samples
• The photographic imaging operator of single lens camera can thus be wriben as:
Single lens camera: ray-space analysis
Image sensor
Single lens camera: ray-space analysis
Aperture
Scene Main lens
Image sensor
Aperture
Scene Main lens
OpGcal axis
x
F
u
OpGcal axis
x Focal plane
F
u Focal plane
u x
Ray-space diagram
Single lens camera: ray-space analysis
Image sensor
Single lens camera: ray-space analysis
Aperture
Scene Main lens
Image sensor
Aperture
OpGcal axis
x
F
u
u x
Ray-space diagram
OpGcal axis
x Focal plane
Scene Main lens
F
u Focal plane
u x
Ray-space diagram
Single lens camera: ray-space analysis
Image sensor
Single lens camera: ray-space analysis
Aperture
Scene
Wide aperture
Image sensor
Main lens
Scene Main lens
OpGcal axis
x
F
u
OpGcal axis
x Focal plane
u x
x
x Integral-projection
Image samples
Ray-space diagram
Single lens camera: ray-space analysis
Narrow aperture
Image sensor
Scene
Narrow aperture
Image sensor
Main lens
F
u
u
Depth-of-field Focal plane
x
x Integral-projection
Scene Main lens
OpGcal axis
Ray-space diagram
Image samples
Ray-space diagram
Single lens camera: ray-space analysis
x
Depth-of-field Focal plane
u
u
x Integral-projection
F
OpGcal axis
x
F
u
u
Depth-of-field Focal plane
x
x Integral-projection
Image samples
Image samples
Ray-space diagram Can we actually capture the light field itself?
Beyond single lens camera
u
• Part 1: Light, op/cs & light-field concepts • Part 2: Tradi/onal camera technology
x
• Part 3: Light field camera technology • Part 4: Sampling & processing the light field
Beyond single lens camera
u
x Integral-projection
Light field
Beyond single lens camera
?
u
x
Light field
• How to acquire directly the light-field?
? x
?
Light field
• How to acquire directly the light-field? • What is an efficient image generaGon process?
Image samples
Beyond single lens camera
u
Beyond single lens camera
?
?
x
u
?
Light field cameras
x
x
Light field
Image samples
?
x
Light field
Image samples
• How to acquire directly the light-field? • What is an efficient image generaGon process? • What advantages can be gained in terms of rendered images?
• How to acquire directly the light-field? • What is an efficient image generaGon process? • What advantages can be gained in terms of rendered images?
Beyond single lens camera
Beyond single lens camera
u Light field cameras
x
Fourier Slice Photograph Theorem
Light field
• How to acquire directly the light-field? • What is an efficient image generaGon process? • What advantages can be gained in terms of rendered images?
u
?
Light field cameras
x
x
Image samples
?
Fourier Slice Photograph Theorem
Light field
• How to acquire directly the light-field? • What is an efficient image generaGon process? • What advantages can be gained in terms of rendered images?
• • • • •
SyntheGc refocusing, SyntheGc aperture Depth map, Parallax images, … x
Image samples
Single lens camera vs. Light field camera
Image sensor
Aperture
Light field camera
Scene Main lens OpGcal axis
Same user-seNngs as in single lens cameras: • Aperture • Focus • Zoom (if available) • Exposure Gme • Sensor sensiGvity (ISO)
Focal plane Image Micro-lens sensor array (mla)
Aperture
Scene Main lens
Image Micro-lens sensor array (mla)
Aperture
Scene Main lens
OpGcal axis
OpGcal axis
Focal plane
Micro-lens array and light field camera
Focal plane
Light field camera
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
Microlens array (Ng06)
Focal plane
Lytro
Raytrix
Light field camera
Image sensor
Light field camera
Micro-lens array (mla)
Scene
Aperture
Image sensor
Main lens
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
OpGcal axis
Focal plane
Focal plane
x
f
u' pixel
Light field camera
Light field camera ANGULAR RESOLUTION
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
Focal plane
x
f
u' pixel
du’
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
Focal plane
Light field camera
Image sensor
Light field camera
Micro-lens array (mla)
Scene
Aperture
Image sensor
Main lens
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
OpGcal axis
Focal plane
Focal plane
x
f
u' pixel
Light field camera
Light field camera: ray-space analysis SPATIAL RESOLUTION
Image sensor
Micro-lens array (mla)
Scene
Aperture
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
Main lens
OpGcal axis
OpGcal axis
x
u
u
Focal plane
Focal plane
dx
f pixel
x u'
dx’
x
x u'
Light field camera: ray-space analysis
Image sensor
Micro-lens array (mla)
Light field camera: ray-space analysis
Scene
Aperture Main lens
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
x
OpGcal axis
x
u
u
Focal plane
x
u
u
Focal plane
x
x
x
u'
Light field camera: ray-space analysis
Image sensor
Micro-lens array (mla)
u'
Light field camera: ray-space analysis
Scene
Aperture Main lens
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
x
x
u
u
Focal plane
x
OpGcal axis
x u'
u
u
Focal plane
x
x u'
Light field camera: ray-space analysis
Image sensor
Micro-lens array (mla)
Light field camera: ray-space analysis
Scene
Aperture Main lens
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
OpGcal axis
x
OpGcal axis
x
u
u
u
Focal plane
u
dx
Focal plane
du x
x
x
x
u'
Light-field camera: ray-space analysis
Image sensor
Micro-lens array (mla)
u'
Light-field camera: ray-space analysis
Scene
Aperture Main lens
Image sensor
Micro-lens array (mla)
Main lens
OpGcal axis
x u
Focal plane
x
OpGcal axis
x
u
Scene
Aperture
u
u Focal plane
x
Light-field camera: ray-space analysis
Image sensor
Micro-lens array (mla)
Light-field camera: ray-space analysis
Scene
Aperture Main lens
Image sensor
Micro-lens array (mla)
Main lens
OpGcal axis
x u
OpGcal axis
x
u Focal plane
u
x
u Focal plane
x
Light-field camera: ray-space analysis
Image sensor
Micro-lens array (mla)
Scene
Aperture Main lens
• Part 1: Light, op/cs & light-field concepts OpGcal axis
x u
u Focal plane
• Part 2: Tradi/onal camera technology • Part 3: Light field camera technology • Part 4: Sampling & processing the light field
x
Scene
Aperture
Digital Photographic Imaging
Digital Photographic Imaging
u
u
F
F Image focused on “green” plane
Image focused on “green” plane
?
x
Light field
x Image samples
Light field
Digital Photographic Imaging
Digital Photographic Imaging
u
u
αF, α1 Image focused on “blue” plane
x
x
x Image samples
Light field
Shear + Integral projection
Image focused on “blue” plane x Image samples
Light field
Imaging Operator of a light field camera
Imaging Operator of a light field camera
• Let be the shear operator that propagates the light-field at :
• Let be the shear operator that propagates the light-field at :
• The photographic imaging operator of a light field camera can thus be wriben as:
Imaging Operator of a light field camera
Fourier and Slicing Operators
• Let be the shear operator that propagates the light-field at :
• Let be the N-dimensional Fourier transform operator of a func0on and let be its inverse:
• The photographic imaging operator of a light field camera can thus be wriben as:
• How about in the Fourier domain?
Fourier and Slicing Operators
Classical Fourier Slice Theorem: illustra/on
u
u
u • Let be the N-dimensional Fourier transform operator of a func0on and let be its inverse:
• Let be the slicing operator that reduces an N-dimensional func0on down to an Mdimensional one by sejng to zero the last N-M dimensions:
x
x
Fourier Transform
Fourier Transform
ku
ku
kx
Classical Fourier Slice Theorem: illustra/on 2D signal
x
Fourier Transform ku
kx
kx
Classical Fourier Slice Theorem: illustra/on 1D signal
u
x
2D signal
1D signal
u
x
x
x
kx
kx'
ku
Classical Fourier Slice Theorem: illustra/on 2D signal
Classical Fourier Slice Theorem: illustra/on 1D signal
u
ku
1D signal
u
x
x
x
x
kx
kx'
ku
kx
kx'
Classical Fourier Slice Theorem: illustra/on 2D signal
2D signal
The Generalized Fourier Slice Theorem 1D signal
u
x
x
kx
kx'
ku
What about 4D signals? And shearing operator?
Let f be an N-dimensional func0on. Then: - changing the basis of f, - integral-projec0ng it down to M dimensions and - applying the Fourier transform is equivalent to: - applying the Fourier transform to f, - changing basis with the normalized inversed transpose of the original basis and - slicing it down to M dimensions.
The Generalized Fourier Slice Theorem
The Fourier Slice Photograph Theorem [Ng06]
Let f be an N-dimensional func0on. Then: - changing the basis of f, - integral-projec0ng it down to M dimensions and - applying the Fourier transform is equivalent to: - applying the Fourier transform to f, - changing basis with the normalized inversed transpose of the original basis and - slicing it down to M dimensions.
• Let recall the photographic operator of a light field camera:
The Fourier Slice Photograph Theorem [Ng06]
The Fourier Slice Photograph Theorem [Ng06]
• Let recall the photographic operator of a light field camera:
• Let recall the photographic operator of a light field camera:
The Fourier Slice Photograph Theorem [Ng06]
The Fourier Slice Photograph Theorem [Ng06]
• Let recall the photographic operator of a light field camera:
A Photograph is the inverse 2D Fourier transform of a dilated 2D slice in the 4D Fourier transform of the light field
The Fourier Slice Photograph Theorem
The Fourier Slice Photograph Theorem
4D signal
2D signal
u
x
4D signal
2D signal
u
x
x
x
kx
kx'
ku
The Fourier Slice Photograph Theorem 4D signal
The Fourier Slice Photograph Theorem 2D signal
u
4D signal
2D signal
u
x
x
ku
x
x
kx
kx'
ku
kx
kx'
Light-field Cameras
Light-field Cameras
• Focused Plenop/c Camera (Adobe): Image mla sensor
• Focused Plenop/c Camera (Adobe):
Aperture
Scene
Virtual image Image mla plane sensor
Main lens
Aperture
OpGcal axis
x
Scene Main lens OpGcal axis
u
x
u
Light-field Cameras
Light-field Cameras
• Focused Plenop/c Camera (Adobe): Virtual image Image mla plane sensor
• Extended Depth-of-field Plenop/c Camera (Raytrix):
Aperture
Scene
Virtual image Image mla planes sensor
Main lens
Aperture
OpGcal axis
x
x
Virtual image plane
Main lens OpGcal axis
u
Scene
u
Virtual image plane
a
a x
b
x
b
u'
u'
ANGULAR RESOLUTION
SPATIAL RESOLUTION
Light-field Cameras
Light-field Cameras
• Extended Depth-of-field Plenop/c Camera (Raytrix): Virtual image Image mla planes sensor
Aperture
• Light-field technology comparison (from Raytrix website 2011): Scene
Main lens
?
OpGcal axis
x
u yes Virtual image planes
Type 1
Type 2
Type 3
Type 1
Type 2
Type 3
Type 1
x u'
yes
Light field image: Lytro example
Raw 4D light->ield
Light field image: Lytro example
Raw 4D light->ield
Light field image: Lytro example
Raw 4D light->ield
Light field image: Lytro example
Raw 4D light->ield
Light field image: Lytro example
Light field image: digital refocusing
u
x
v
y
pixel
Light field image: digital refocusing
Light field image: perspec/ve shi_
Let view All-in-Focus
Right view
Light field image: perspec/ve shi_
Custom light-field camera
Light field image: depth map
Custom light-field camera
Nikon D800 + Micro-Nikkor 105mm f/2.8G
XY-slide
XY-slide Schneider Fine Art XXL Coppal #3 + Sinar lens board Focal length
X-slide
X-slide Focal plane
PainGng OpGcal axis
Acquiring the 4D light field of a painGng
Custom light-field camera Focal length Focal plane
Camera posiGons
OpGcal axis
Sub-aperture images
Region of interest
Custom light-field camera
Custom light-field camera Focal length
Camera posiGons
Focal length Focal plane
Camera posiGons
OpGcal axis
Light rays from ith pixel from each sub-aperture image
Focal plane
OpGcal axis
Light rays from ith pixel from each sub-aperture image Light rays from jth pixel from each sub-aperture image
Custom light-field camera
Custom light-field camera Focal length
Camera posiGons
Focal plane
OpGcal axis Focal plane
PainGng
Focal plane àConjugate pinhole cameras PainGng
Acquiring the 4D light field of a painGng
Sub-aperture image view of acquired light field (major: UV ; minor: XY) 32x32x512x512
Acquiring the 4D light field of a painGng (resoluGon: 512x512x32x32)
Microlens view of acquired light field (major: XY ; minor: UV) 512x512x32x32
Sub-aperture image view of acquired light field (major: UV ; minor: XY) 32x32x512x512
Microlens view of acquired light field (major: XY ; minor: UV) 512x512x32x32
Depth map from acquired light field
What to remember
• Light field processing is at the convergence of opGcs and image processing. • Ray tracing offers a convenient framework to manipulate light field. • OpGcal elements such as thin lens and free space shear the light field. • The light field is 4D signal and an image is one 2D integral-projecGon of it. • Light field cameras acquire both spaGal and angular informaGon of light rays. • From a light field, images with different focus can be generated ater acquisiGon. • Working in the Fourier domain is computaGonally more efficient. • The Fourier Slice Theorem can be applied to the light field for image rendering.