generalized mosaics yoav y. schechner, shree nayar department of computer science columbia...

30
Generalized Mosaics Yoav Y. Schechner , Shree Nayar Department of Computer Science Columbia University

Upload: caitlin-wilkinson

Post on 15-Jan-2016

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Generalized Mosaics

Yoav Y. Schechner,

Shree Nayar

Department of Computer Science

Columbia University

Page 2: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Mosaics

Processing

Page 3: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Redundant Measurements

Generalized Mosaicing: Yoav Schechner and Shree Nayar

Page 4: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Scanning with Less Redundancy

Mantis Shrimp

TakataC

Different rows = Different optical characteristics

Page 5: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Generalized Mosaicing

Field of View

BrightnessDynamic Range Spectrum

Depth of Field

Polarization

camera

Spatially varying

filter

Schechner, Nayar, Generalized mosaics

Page 6: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Mosaic + High Dynamic Range

88 - 18,794

Page 7: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Fusion of Measurements

)(Mask),(),(Raw xyxIyxk

1

22 1

k kI

I

k k

k

I

III

22

Maximum-Likelihood Solution

),( ),( )(Mask

),(RawyxIyxI

x

yxkk

k

Schechner, Nayar, Generalized mosaics

Page 8: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

8 bit camera Dynamic range as 16 bits

Generalized Mosaicing: Yoav Schechner and Shree Nayar

8 bits (almost)(pixels)

-2

-4

-6

-8 x

log M2

Log of the mask

==1 10-61 10-2

Page 9: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

M

x

average row

frames y

frameyxx ),(Raw )(Mask

),(Mask),(),(Raw yxyxIyxframe

Mask Self-CalibrationUnknown filter (vignetting)

consistencyconstraints

0Raw)(Mask Raw)(Mask1221 xx

MaskttMaskmin AA

Consistency, Smoothness A

Schechner, Nayar, Generalized mosaics

Page 10: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Image Registration

2pair ingcorrespondeach

pixels all

2total DIFFDIFF

2total DIFF:Minimize

2pair ingcorrespondeach

pixels all

2total DIFFDIFF

Schechner, Nayar, Generalized mosaics

Page 11: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

)(Mask),(),(Raw xyxIyxk

2pair ingcorrespondeach

pixels all

2total

DIFFDIFF :Minimize

Bias towards “no motion”

Schechner, Nayar, Generalized mosaics

Image Registration

Page 12: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

)(Mask),(),(Raw xyxIyxk

),( )(Mask

),(RawyxI

x

yxk

k

Raw

Mask

Uncertainty

),( yxIk

1

22

22 1 where ,

k kk k

k

II

I

III

2

22

2

2

21

12pair pixel

III

III

DIFF

Schechner, Nayar, Generalized mosaics

Image Registration

Page 13: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Registration: Standard “Coarse to Fine”

5 pixels

10 pix 9 pix

18 pix 19 pix

4 pixels

2 pixelsmotion of

However, we need: multiscale UNCERTAINTIES 2

22

2

2

21

12pair pixel

III

III

DIFF

Schechner, Nayar, Generalized mosaics

Page 14: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Maximum-Likelihood Pyramid At each level L , for each pixel

LL II , estimate

,

),( ),(yx

yxIyxwI

,

),( yx

yxw

,

),( 2 yx

yxwI-1

),(

),(),(yxI

yxayxw

2

Schechner, Nayar, Generalized mosaics

Page 15: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Max-Likelihood Pyramid

Gaussian Pyramid1

22

22

)(

1

)(

)( ,

xx xII

xI

xIII

x

)(xg

)1(I )2(I )3(I )4(I )5(I

x

xIxgI )()(

x )1(I )2(I )3(I )4(I )5(I )1(I )2(I )3(I )4(I )5(I

ML estimation

x

xIxnI )()(

)(

)()(2 xI

xgx

1][ )(

xn

1

2 )(

x

xI

Page 16: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

log I

Inte

nsit

y ra

nge

Spatial range x

Generalized Mosaic

x

log I

Inte

nsit

y ra

nge

Spatial range

frame

fram

e

Mosaic

Schechner, Nayar, Generalized mosaics

Page 17: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Generalized Mosaicing: Yoav Schechner and Shree Nayar

spectral

y

x

Variable Spectral Filter

Page 18: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

MMuullttii SSppeeccttrraall Mosaic

500 700400 600 500 700400 600 500 700400 600

Generalized Mosaicing: Yoav Schechner and Shree Nayar

Page 19: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Generalized Mosaicing: Yoav Schechner and Shree Nayar

Rendering : Any Illumination

sunset fluorescent

Page 20: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Halogeny

x

Illumination at a Glance

Fluorescenty

x

Illuminant spectrum

Extra information

Page 21: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Spatially Varying Polarizer

Polarization Mosaicing: Yoav Schechner and Shree Nayar

Page 22: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

0

0.2

0.4

0.6

0.8

1

-30

60

90

30

0

-60

-90

o

o

o

o

o

o

polarizance

polarizing angle

o

Polarization Mosaicing: Yoav Schechner and Shree Nayar

transmittance

Spatially Varying Polarizer

Page 23: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Raw images

Polarization Mosaicing: Yoav Schechner and Shree Nayar

Page 24: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Polarization Mosaic

Schechner Shamir Kiryati

JOSA-A 2000

reflected structure

transmitted painting

Page 25: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

=

y

xDepth

Insert a prism behind the lens

Spatially Varying Focus

Generalized Mosaicing: Yoav Schechner and Shree Nayar

Page 26: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University
Page 27: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

All FocusedAll Focused Mosaic

Generalized Mosaicing: Yoav Schechner and Shree Nayar

… and a Depth Map

Page 28: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

What else?

Generalized Mosaicing: Yoav Schechner and Shree Nayar

Dynamic Range & Spectrum & Polarization etc.

Multi-dimensional Mosaics - Simultaneously

Page 29: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

Signal undergoing LPF

“Band limited” signal

Minimizing aliasing

Nyquist sampling rate

111111111111111

1?10111001??

Sampling Criteria

Generalized Mosaicing: Yoav Schechner and Shree Nayar

Number of samples per scene point

resolution

length Lbandwidth B

aperture D

22 )/()/(

2

BLD

M2

log11 M

1011100101

1011100???

M # samples

1 11/2 21/4 3

92561

101110????

Page 30: Generalized Mosaics Yoav Y. Schechner, Shree Nayar Department of Computer Science Columbia University

MosaicGeneralized Mosaic

Spatially varying filter

Wide field of view

SpectrumI

Depth + Focus

Dynamic range88 – 18,794

Polarization