Subjective Image Quality Metrics from The Wave...

Subjective Image Quality Metrics fromThe Wave Aberration

David R. WilliamsWilliam G. Allyn Professor of Medical Optics

Center For Visual ScienceUniversity of Rochester

Commercial Relationship:Bausch and Lomb

Funding:Bausch and Lomb

NSF Science and Technology Center for Adaptive Optics

Collaborators:Univ of Rochester:Li ChenNathan DobleHeidi Hofer

Ian Cox, Bausch and LombRay Applegate, University of HoustonLarry Thibos, Indiana University

Scott MacRaeJason PorterBen SingerRemy Tumbar

How Bad Is This Wave Aberration?

Goal:

To compute a number that captures the subjective effect of the eye’s wave aberration.

Uses:

Assessing severity of the wave aberrationCalculating the best correction

How Bad Is This Wave Aberration?

RMS Wavefront Error = 0.87 µm

Defocusrms = 0.5 µm

Spherical Aberration

rms = 0.16 µm

Defocus andSpherical Aberration

rms = 0.52 µm

Some Aberrations Interact Strongly in Image Blur

Zernike Modes

trefoil coma coma trefoil

quadrafoil secondary spherical

2nd

3rd

4th

5th

radialorder

secondary quadrafoil

pentafoil secondary secondarysecondary pentafoilsecondary

defocusastigmatism astigmatism

astigmatism astigmatism

trefoiltrefoil comacoma

Z20Z 2

-2 Z22

Z 3-1 Z3

1Z 3-3 Z3

3

Z40 Z4

2Z 4-2 Z4

4Z 4-4

Z51Z 5

-1 Z53Z 5

-3 Z55Z 5

-5

Lower OrderAberrations

Higher OrderAberrations

Zernike Modes



2nd

3rd

4th

5th

radialorder






Z20Z 2

-2 Z22

Z 3-1 Z3

1Z 3-3 Z3

3

Z40 Z4

2Z 4-2 Z4

4Z 4-4

Z51Z 5

-1 Z53Z 5

-3 Z55Z 5

-5

Lower Order Aberrations


Zernike Modes



2nd

3rd

4th

5th

radialorder






Z20Z 2

-2 Z22

Z 3-1 Z3

1Z 3-3 Z3

3

Z40 Z4

2Z 4-2 Z4

4Z 4-4

Z51Z 5

-1 Z53Z 5

-3 Z55Z 5

-5

Lower OrderAberrations


There are strong interactions between Zernike Modes.

Therefore,Decomposing the wave aberration

into Zernike modes is not the best way to evaluate the subjective impact of

the wave aberration

Point Spread Function

How Bad is This Wave Aberration?

Wave Aberration

Use Retinal Image Quality, Not the Wave Aberration

Aberrations in Lens and Cornea Distort Wave front

Principle of Adaptive Optics

Deformable MirrorCorrects Wave front

Sharp Image in Camera

Wave front SensorMeasures Wave front

Adaptive Optics Sharpens the Eye’s Point Spread Function

Adaptive Optics Can Create Wave Aberrations

Wave Aberration After AO correction With coma added

(Subject: ND �)

Convoluted retinal image

Wave aberrations from Lasik postOp patient

Wave aberrations created by adaptive optics(With real eye, JP)

Blur Matching

• Lots of Sharp Edges• Edges At All Orientations• Seen Through Adaptive

Optics

550 nm, 1 deg, 6 mm pupil

Binary Noise Stimulus

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

DefocusPatient wave aberration Stimulus

The subject adjusted the amplitude of defocus so that thesubjective blur matched that of the patient wave aberration.

Blur Matching of Patient Wave Aberrations

Zernike mode Zernike mode

Am

plitu

de(µ

m)

Def

ocus

(D)

Blur Matching Controls for Neural Differences between

Patients

Using Multiple Subjects Controls for Neural Adaptation

Acuity does not always capture thesubjective quality of vision

Equal Acuity But Different Subjective Image Quality

Compare Subject Matches with Matches Made Using Three Different Metrics

Wavefront RMSStrehl Ratio

Neural Sharpness

Wave Aberration RMS

-3 -2 -1 0 1 2 3-1

0

1

2

3

Am

plitu

de (µ

m)

Aperture(mm)

Big RMS

Small RMS

Big RMS Small RMS

R2 = 0.4627

0

0.2

0.4

0.6

0.8

1

1.2

0 0.2 0.4 0.6 0.8 1 1.2

RMS Metric

Met

ric m

atch

es (D

)

Subject Matches (D)

Strehl Ratio of Point Spread Function

Diffraction-limited PSF(Perfect Eye)

Hdl

Heye

Actual PSF(Aberrated Eye)

Strehl Ratio = H eye

H dl

C. of Austin Roorda

R2 = 0.4902

0

0.2

0.4

0.6

0.8

1

1.2

0 0.2 0.4 0.6 0.8 1 1.2

Strehl Ratio MetricM

etri

c mat

ches

(D)

Subject Matches (D)

( ) SubjectiveImage

Quality Point Spread

FunctionGaussian Neural Blur

σ = 0.8’

Σ x =

A Simple, Biologically-Plausible Metric for Subjective Image Quality

R2 = 0.7034

0

0.2

0.4

0.6

0.8

1

1.2

0 0.2 0.4 0.6 0.8 1 1.2

Neural Sharpness Metric

Met

ric m

atch

es (D

)

Experiment (D)

R2 = 0.7034

0

0.2

0.4

0.6

0.8

1

1.2

0 0.2 0.4 0.6 0.8 1 1.2

Neural Sharpness

Met

ric m

atch

es (D

)R2 = 0.4627

0

0.2

0.4

0.6

0.8

1

1.2

0 0.2 0.4 0.6 0.8 1 1.2

Met

ric M

atch

es (D

)

Subject Matches (D)

Wavefront RMS

Collaboration to Identify the Optimum Image Quality Metric

Ray Applegate, University of Houston:Effectiveness of Image Quality Metrics in PredictingVisual Acuity with Convolution Simulations

David Williams, University of RochesterEffectiveness of Image Metrics in predictingSubjective Image Quality with Adaptive Optics

Larry Thibos, Indiana University:Effectiveness of Image Quality Metricsin Predicting Visual Acuity in the Population

defocus

met

ric

valu

e

maximum

Optimizing refraction with an image quality metric

error

subjective

search in3-D space

Guirao and Williams (2003)

Conclusions• Generating blur with adaptive optics leads to a robust metric for correcting vision.

• It is hard to estimate subjective blur from the wave aberration. Zernike Decomposition doesn’t help much.

• The point spread function combined with a biologically plausible model of neural blur is better.

• Standards for optimizing correction from wavefront are in the works

Subjective Image Quality Metrics from The Wave...

Documents

Transcript of Subjective Image Quality Metrics from The Wave...