lab 5

• progress?
• questions?

today's schedule

• Recap
• Coded exposures
• Motion Invariant Photography
• Coded rolling shutter

recap

coded illumination

several illumination coding strategies

temporal coding

directional coding

spatial coding

assisted stereo

source of blur

image Formation

Deblurring is a deconvolution

deconvolution: 2 types

ill-posed problem

focus blur

aperture and PSF

aperture and PSF

• Depth-dependent PSF
• Engineer PSF to be depth invariant -> deconvolution is easier


• Circular apperture is not (well) invertible
• Engineer PSF to be broadband (flat Fourier magnitudes) -> well posed (Coded Appertures)
• Coded apperture for depth recovery
• Coded apperture based on perception

focal sweep

focal sweep

Extended Depth of Field (EDOF)

coded apertures

coded apertures

coded apertures

1D intuition

prior

depth adn deblurring

engineered PSF

coded exposures

motion blur

blur is 60 pixels

blur is 235 pixels

blur is 118 pixels

previous work

previous work

focus on PSF estimation

focus on capture

The convolution can also be written a a linear system.

$Ax = b$

?

coding motion

last week...

how about making motion PSFs motion invariant?

why motion deblurring is hard?

• Need to know blur kernel (motion velocity)
• Need to segment image
• Information loss (reduced signal to noise ratio)
• Coded exposures?

Doesn’t address the first two challenges

motion blur problems

motion blur problems

[Levin et al.2008]

• To reduce motion blur, increase it!

- move camera as picture is taken

• Makes blur invariant to motion

- can be removed with spatially uniform deconvolution

- kernel is known (no need to estimate motion)

- kernel identical over the image (no need to segment)

• Makes blur easy to invert

controlling motion blur

controlling motion blur

controlling motion blur

controlling motion blur

controlling motion blur

controlling motion blur

parabolic sweep

motion invatiant blur

motion invatiant blur

motion invatiant blur

space-time volume

space-time volume

camera integration

shearing

shearing

parabolic curve:shear invariant

parabolic curve:shear invariant

parabolic curve:shear invariant

parabolic curve:shear invariant

parabolic curve:shear invariant

parabolic curve:shear invariant

parabolic curve:shear invariant

parabolic curve:shear invariant

deblurring and information loss

space-time Fourier domain

space-time Fourier domain

space-time Fourier domain

space-time Fourier domain

space-time Fourier domain

static camera

flutter shutter

parabolic camera

flutter shutter

parabolic camera

cameras and information preservation

comparing camera reconstruction

hardware construction

linear rail

linear rail

no perfect linearity

1D motion violation: forward

1D motion violation: stand-up

1D motion violation: rotation

issues

• Spatial shift- but does not affect visual quality in deconvolution
• Parabola tail clipping: not exactly the same blur
• Motion boundaries break the convolution model
• Assumes: Object motion horizontal
• Object motion linear up to 1st order approximation

• Camera moved during exposure, parabolic displacement
• Blur invariant to motion:

- Same over all image (no need to segment)

- Known in advance (no kernel identification)

• Easy to invert (near optimal frequency response)
• For 1D motion

- Somewhat robust to 1D motion violation

- Future work: 2D extensions

?

a couple more papers

coded rolling shutter

Video from Coded Exposure

lab due next week!

SIGGRAPH Volunteers

http://franchomelendez.com/Uwr/teaching/COMPHO/_LECTURES/L8/coded_exposures.html

franchomelendez@cs.uni.wroc.pl

credits and references and aditional readings

These slides have been prepared with materials, slides, and discussions from the following.

• Gordon Wetzstein
• Anat Levin
• Ramesh Raskar