resolution loss without optical blur tali treibitz alex golts yoav y. schechner technion, israel 1

49
Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion , Israel 1

Upload: junior-gibbs

Post on 17-Jan-2016

214 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Resolution Loss without

Optical Blur

Tali Treibitz

Alex Golts

Yoav Y. Schechner

Technion , Israel

1

Page 2: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

14

airlightA

( , ) ( , ) ( ) ( , )D

I x y L x y t z A x y

object

0

1

z

ze

scatteringdirect transmission

D

object radiance

Lobject

total intensity

I

A

0 z

1 ze

Schechner, Narasimhan, Nayar

Page 3: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Haze

object( ) ( ) ( ) ( )I l t A x x x x

airlightobject transmittance

Schechner et al., Applied Optics ‘03

15

Page 4: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Pointwise Degradations

object( )I lx

object

( )x t ( )x A

pointwise attenuation:

vignetting

atmosphere attenuation

additive component:

reflection

glare

path radiance

( )x n

noise

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

17

Page 5: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Pointwise Degradations

object( )I lx

object

( )x t ( )x A

pointwise attenuation

( )x n

noise

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

reduce SNR even if known

additive (positive) bias

18

Page 6: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Noise: Object size matters?19

Page 7: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Noise: Object size matters?

0.5

20

Page 8: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

1

Noise: Object size matters?21

Page 9: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

1.5

Noise: Object size matters?22

Page 10: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

depends on:

noise level

object background intensity difference

object size

quantify this dependency!

Prior art: resolution limits due to optical blur

here: no optical blur

Visibility Under Noise23

Page 11: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Previous criteria• Is there something there?• Is it a tank?• What type is it?

Johnson charts:

identification recognition orientation detection

7 3.5 1.2 0.75 tank

minimum line pairs for 50% success

Page 12: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

NIIRS- National Image Interpretability Rating Scales

Identify the wing configuration (e.g., straight, swept, delta) of all large aircraft (e.g., 707, CONCORD, BEAR, BLACKJACK). ..

Detect large hangars at airfields. Detect large static radars (e.g., AN/FPS-85, COBRA DANE, PECHORA, HENHOUSE), Detect military training areas...

Detect a medium-sized port facility and/or distinguish between taxi-ways and runways at a large airfield.

Interpretability of the imagery is precluded by obscuration, degradation, or very poor resolution

Identify small light-toned ceramic insulators that connect wires of an antenna. Identify vehicle registration numbers (VRN) on trucks. Identify screws and bolts on missile components. ..

0

1

2

3

9

Page 13: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

pattern visible

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

24

Page 14: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Where is the Cutoff?

0.1 0.5

10

0.5

pattern visible

pattern invisible calculated analytically!

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

Input SNR

noise

| ( ) |

S u

25

(frequency)u

Page 15: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

0.1 0.5

10

0.5

(frequency)

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

u

26

Cutoff Per Success Rate

success rate 50%

cutoffu

1

Input SNR

noise

| ( ) |

S u

Page 16: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Noise Suppression in the HVS

Theoretical Neuroscience, Dayan & Abbott

frequency (cycles/degree)

response of receptive field

low noise

high noise

We derive: fundamental analytical model

Model: simple linear denoising

not a denoising method

28

Page 17: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

SNR Improvement by Averaging

WH

/W

signal

noise

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

output

input

SNRC(u,W)=

SNR- SNR change after averaging

29

Page 18: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Different Sizes of Windows

too big for signal too small for noise

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

30

Page 19: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Averaging by Optimal Window

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

31

Page 20: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

SNR Improvement by Averaging

outputmax input( ) SNR u SNR

depends on frequency!

u

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

max ( )C u

same plot for a Gaussian filter

32

Page 21: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Output SNR

0.1 0.5

6.5

0.32

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

cutoffu

output 1SNR output 0.5SNR

Input SNR

noise

| ( ) |

S u

(frequency)u

1

33

Page 22: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Cutoff Per Success Rate

0.1 0.5

0.32

success rate 70%

success rate 40%

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

Input SNR

(frequency)ucutoffu6.5

noise

| ( ) |

S u

34

Page 23: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Vision Success is Probabilistic

SNR=2/3

SNR determines chances of visibility

visible invisible

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

35

Page 24: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Success within a Confidence Interval

( ) prob N S xSN

68%

0 22

- success rate

SNRTreibitz & Schechner, Recovery Limits in Pointwise Degradations

0.68

Object is visible if ( )N Sx

…depends on SNR and..

( )xN

36

Page 25: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

25

Success within a Confidence Interval

- success rate

SNRTreibitz & Schechner, Recovery Limits in Pointwise Degradations

what is the probability for correct detection?

depends on SNR object pixel

background pixel

visibility is kept if edge keeps sign

background object ( ) ( ) x x prob N N S

%(sign kept) - %(wrong sign)

noisyclear

0.5

Page 26: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

0.1 0.5

6.5

0.32

system input SNR

cutoffu

Determining Resolution Limits

cutoff for ρ=70% success

frequency

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

noise

| ( ) |

S u

37

Page 27: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Pointwise Degradations

object( )I lx ( )x t ( )x A

pointwise attenuation:

vignetting

atmosphere attenuation

additive component:

reflection

glare

haze

( )x n

noise

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

38

Page 28: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Noise Model

2 2 / I g

2 200 I

Nikon D100

2

photon noise dominates

39

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

Page 29: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Detector (pixel)

50% quantum efficiency

Schechner

Photon (shot) Noise9

Electrons

Photon

or

e{nothing

either

Page 30: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

50% quantum efficiency

Schechner

10

Photons

Electrons

e{nothing

e e

either

Photon (shot) Noise

Page 31: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

object background| | , ( )

SSNR = S l l t x

SNR per size (frequency)

object background| | ,

SSNR = S l l

objectl

backgroundl

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

backgroundlobjectl

41

Page 32: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Resolution Limits in Haze

distance [km]

limit due to pixel size

limit due to atmosphere

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

minimal visible object size[m]

cutoffu

reciprocal to

42

Page 33: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

0.1 0.5

10

0.5

(frequency)

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

u

43

Cutoff Per Success Rate

success rate 50%

cutoffu

1

Input SNR

noise

| ( ) |

S u

Page 34: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Haze in the Galilee

Treibitz & Schechner, Recovery Limits in Pointwise Degradations

average of 50 framesraw frame

limit due to noise and not blur

44

Page 35: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

What now?

What are the reconstruction limits?

What is the minimal detectable object size?

What camera noise properties are acceptable for detection?

45

Page 36: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Imaging in Haze

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

46

Page 37: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

Haze Through a Polarizer

best polarized image

47

Page 38: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

single frame- used by photographers

Haze Through a Polarizer

increased exposure time

best polarized image

48

Page 39: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

two frames- Schechner et al.

Dehazing using a Polarizer

post-processing 2 frames

best polarized image

worst polarized image

49

Page 40: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

goal: object detection local contrast stretch- OK

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

Is it worth using a polarizer?

unpolarized image

best polarized

image

post-processing 2 frames

rarely!under the constraint of equal acquisition time

50

Page 41: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

degree of polarization

( , ) 1( ) ( , )

2 2

D x y pI x, y A x ymin

0,1p

Using a Single Polarized Image

Best polarized image

I min

51

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

Page 42: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

SNR ComparisonIunpolarized Ibest

52

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

Page 43: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

p

A Single Saturated Frame

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

SNRpolarized

SNRunpolarized

maximal value in nature

53

Page 44: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

SNR ComparisonIunpolarized Ibest

equal acquisition time

technical details

in the paper

acquisition time = exposure time X number of frames

54

Page 45: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

Same Total Acquisition Time

p<0.4in our experiments

SNRpolarized

SNRunpolarized

maximal value in nature

p

55

Page 46: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Experiment

Wide field of view

bestI

average of 2 frames

unpolarizedI

Treibitz & Schechner, Polarization- Beneficial for Visibility Enhancement?

same total acquisition time

56

Page 47: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

dehazing

tbest tworst

1 p

p

1 p

p

SNR ComparisonIunpolarized

technical details

in the paper

optimal exposures

equal acquisition time

57

Page 48: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

Advantages of Polarization

distance map

• contrast stretch in non-uniform distances

• restoring color• compensating for attenuation

58

Page 49: Resolution Loss without Optical Blur Tali Treibitz Alex Golts Yoav Y. Schechner Technion, Israel 1

• Freq cutoff – due to noise – without imaging blur

• Relation between cutoff and success rate

• Application: limits in pointwise degradations

Limits in Pointwise Degradations

• Case study of performance trade-offs

59