partially coherent ambiguity functions for depth-variant point spread function design roarke...

30
Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1 PIERS Marrakesh, 3/20/11 1 MIT Media Lab 2 MIT Department of Mechanical Engineering

Upload: duane-ray

Post on 16-Jan-2016

223 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design

Roarke Horstmeyer1

Se Baek Oh2

Otkrist Gupta1

Ramesh Raskar1

PIERS Marrakesh, 3/20/11

1MIT Media Lab2MIT Department of Mechanical Engineering

Page 2: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Conventional camera PSF measurement:

Conventional PSF Blur

defocus

Circular Aperture

I(z0) I(z1) I(z2)

z0 z1 z2

f

Pt.

So

urce

Problem Statement

Page 3: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

3

Design apertures for specific imaging tasks

Determine mask

Aperture mask:amplitude/phase

defocus

Desired set of PSFsI(z0), I(z1), I(z2)

z0 z1 z2

Problem Statement

Page 4: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

ExamplesDe

focu

s

PSFs: Depth-Invariant Rotating Arbitrary

Cubic phase Gauss-Laguerre modes Iterative Design

?

Page 5: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

PSF Design Similar to Phase Recovery

d

z0 z1 z2

Diff

racti

ve

Elem

ent

f

z0 z1 z2

Aper

ture

El

emen

t

3D PSF Design Measurement for phase recovery

I(z0), I(z1), I(z2) I(z0), I(z1), I(z2)

General Goal: Find (A,ϕ) from multiple intensities

(A,ϕ) (A,ϕ)

Similar to

Models: Fresnel propagation, k-space, light fields, phase space

Page 6: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

I(x1,z1)I(x0,z0)U(A,ϕ)

(a) Phase Retrieval

iterative

iterative

I(xn,zn)I(x0,z0) …

(c) Phase Space Tomography

U(A,ϕ)

simultaneous

I(x1,z1)I(x0,z0)

(d) Mode Selective (proposed)

U(A,ϕ)

simultaneous

iterative

Overview of 3D Design Techniques

I(x1,z1)I(x0,z0)

(b) Transport of Intensity

U(A,ϕ)

simultaneous

Page 7: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

I(x1,z1)I(x0,z0)U(A,ϕ)

(a) Phase Retrieval

iterative

iterative

I(xn,zn)I(x0,z0) …

(c) Phase Space Tomography

U(A,ϕ)

simultaneous

I(x1,z1)I(x0,z0)U(A,ϕ)

simultaneous

iterative

I(x1,z1)I(x0,z0)U(A,ϕ)

simultaneous

Phase space extends nicely to partially coherent design

(b) Transport of Intensity

Overview of 3D Design Techniques

(d) Mode Selective (proposed)

Page 8: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Phase Space Functions

Wigner Distribution (WDF) Ambiguity Function (AF)

AF “easier” than WDFTu, Tamura, Phys. Rev. E 55, 1997

OTF(z1)

OTF(z0)

OTF(-z1)

F-slice

PSF(z0)

PSF(-z1 ) PSF

(z 1)

WDF Projections: PSFs

x

u

u

x'AF Slices: OTFs

Page 9: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

z0 z1

fx

U(x)

Phase Space Camera Model r

Δz

z0I

I

z1

OTFs

Aperture mask

Page 10: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

z0 z1

fx

U(x)

tan(θ0)=0

u

Phase Space Camera Model

Why AF is useful:

z0I

I

z1tan(θ1)=(W20k/π)

OTFs

r

Δz1. Polar display of the OTF

z0

z1

Aperture mask

Page 11: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Aperture mask z0 z1

fx

U(x)

tan(θ0)=0

u

Phase Space Camera Model

Why AF is useful:

z0I

I

z1tan(θ1)=(W20k/π)

OTFs

r

Δz

2. Convert AF to Mutual Intensity: inverse FT, 45° rotation, scale by 2

3. Recovery of U(x) from AF (up to constant Δϕ)

z0

z1

1. Polar display of the OTF

Page 12: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

OTF(z1)

OTF(z0)

PSF OTF Inputs

Output: Desired Aperture Mask, 1D

(1) AF Population

(2) One-time AF Interpolation

OTF(z2)

θn

(5) Optimized AF

u 1

0

u

u

x2

x1

x1

x2

(3) Mutual Intensity J

Rank Constraint

(4) Optimized J

Error Check

1

0

1

0

Iterate &

Page 13: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Rank-constraint on Mutual Intensity, J

= λ1 + λ2 +…+ λ3

x1

x2

x1

x2

Represent J with coherent mode decomposition1

Coherent, orthogonal modes from singular value decomposition

J = UΛVT = Σ λi Ui(x1)Ui*(x2)

Imperfect J guess: Many coherent modes

Assume: J symmetric, nxn λi = Singluar ValuesUi orthogonal to Uj for all i≠ji=1

n

J

1 E. Wolf, JOSA 72 (3), 1982

Page 14: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

= λ1 + λ2 +…+ λ3

1st Mode: Coherent

x1

x2

x1

x2

PSF = response to a point source: restricted to 1 mode

Rank-constraint on Mutual Intensity, J

Jest = λ1 U1(x1)U1*(x2)

J

Represent J with coherent mode decomposition1

Coherent, orthogonal modes from singular value decomposition

1 E. Wolf, JOSA 72 (3), 1982

Page 15: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Ground Truth AF Reconstructed AF Computed Phase Mask

Ground Truth and Reconstructed OTFs

W20=0 W20=λ/2 W20=λ

u

xʹ xʹ xʹ

u

Reconstruction Example: Cubic Phase Mask

Example aperture mask function: exp(jαx3), α=40, 20 iterations

Page 16: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Simulation

Experiment

Amplitude

One (fixed) mask

Simple Example: Arbitrary InputInput

I1(x) I2(x) I3(x) z1=50mm z2=50.1mm

Rank-1 constraint

25μ – resolution, 1cm2 binary mask in 50mm f/1.8 Nikkor, 200μ pinhole @ z=4m

z3=50.2mm

50μ

Page 17: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

No Constraints on (A,ϕ)

Ph

ase

(rad

.)

x (cm)

Am

plitud

e (AU

)

Amplitude-onlyconstraint

Phase-only constraint

Constrained Decompositions

In Experiment: Amplitude-only or Phase-only required

MSE vs. # iterations

MS

E

# of iterations

Aperture mask constraints:

-Varied performance-Algorithm still converges

Page 18: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Keeping More than One Mode

= λ1 + λ2 +…+ λ3

Several Modes: Partially Coherent

x1

x2

x1

x2

-More accurate estimate found with n > 1 modes

J

n = 3: (J - Σ λiUi(x1)Ui(x2))2 = global minimum

Eckert-Young Thm.: 1st n-modes of SVD(J) = optimal rank-n estimate

SVD = Optimal estimate (L2 norm, no prior knowledge)

i=1

3

Page 19: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Simulating Partial Coherence

= λ1 + λ2 +…+ λ3

x1

x2

x1

x2

-More accurate estimate found with n > 1 modes

Multiple modes can be multiplexed over time1,2

JSeveral Modes: Partially Coherent

1 P. De Santis, JOSA 3 (8), 1986, 2 Z. Zhang, private communication, 2011

Spatial Light Modulator:Vary over time

CPU

Page 20: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Simulation

Example: Benefit of Several ModesInput

MSE improvement ~100x(modes contain both A and ϕ)

1 2

3

Display 3 Optimal masks

I1(x) I2(x) I3(x)z1=50mm z2=50.1mm z3=50.2mm

50μ

1cm2 masks

Rank-3 constraint

Experimentally: A,ϕ over time = hard

“Weights”:1 - 12 - .613 - .38

Page 21: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Adding a Constraint to Several Modes

SVD & constrain in separate operations: No convergence

= μ1 + μ2 +μ3

x1

x2

On(J)

x1

x2

General solution: convex optimization

e.g.: amplitude-only, phase-only, spatial1 and coherence constraints

min || J – Σ μiWi Wi* || 2 subject to constraints on W, given n

i=1

n

1 Flewett et al., Optics Letters 34 (14) 2010

W1=? W2=? W3=?

Operation On(J) = find closest n rank-1 outer-products, constrained

Page 22: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Example Constraint: Amplitude-only

Problem: n optimal coherent modes that are real, positive

min || J – WWT || 2 W ≥ 0, real (J = kxk, W=kxn)

Page 23: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Example Constraint: Amplitude-only

Problem: n optimal coherent modes that are real, positive

min || J – WWT || 2 W ≥ 0, real (J = kxk, W=kxn)

Solution: Non-negative matrix factorization1

-e.g. Netflix challenge: low-rank rep. of 0-5 star movie scores

Symmetric NMF: add to update rules (solve for W &H, W≈H)

Note: Optimal “Coherent modes” are no longer orthogonal

Update Rules: Add line 1δ=tiny value, error ~2-5%

1. H=WT

2. W=W.*(HTJ)./((HHT)H+δ)3. H=H.*(WTJ)T./H(WW)+δ)

1Lee and Seung, Nature 401, 2001

Page 24: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

A Simple Example: Multiple Amplitude ModesI1(x) I2(x) I3(x)

1 2

3

1cm2 masks

Amplitude-only, 3 masks

Amp-only masks: Sym. NMF

“Weights”:1 - 12 - .783 - .08

Simulation

Input

MSE improvement ~7x (vs. 1 Amp. mode)

Buildup of a baseline bias…

z1=50mm z2=50.1mm z3=50.2mm

50μ

Page 25: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Conclusion & Future Work

-Phase space functions = intuitive window into 3D PSF design

-Multiple modes (partially coherent) = increased flexibility

-Constrained searches can be achieved w/ convex methods- Amplitude-only: Symmetric NMF- Other constraints: Phase-only (another convex

implementation), coherence length (weighted SVD)- Subtracting modes: J=U1U1

* ± U2U2*±… (take 2 images)

-Current: Initial Experimental tests using an SLM

-Next Step: Find a nice application

Thanks! Questions?

Page 26: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Jpc(x1,x2)

AFpc(x',u)

x1(cm)

x2

u

.5

3 Coherent Modes

1cmMask

x'(cm-1) 5e4-5e4

-.5 .5-.5x(cm)

u

x2

Partially Coherent Reconstruction: 3 Modes

x1(cm) .5-.5

x'(cm-1) 5e4-5e4

Ground Truth Reconstructed

Page 27: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Constraining Several Modes

Apply Constraint: Amplitude-only, Phase-only, prior knowledge, etc.

= λ1 + λ2 +…+ λ3

x1

x2

Sum=No longer optimal (localized constraints will not converge)

individualconstraint

x1

x2

x1

x2

+ λ2 + λ3λ1

SVD(J)

amplitude-only amplitude-only amplitude-only

Example: Amplitude-only maskindividualconstraint

individualconstraint

Page 28: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

A Simple Example: Prior Knowledge

A B

C

3 modes hitting unknown structure

(A=.4, B=.5, C=.7)+

SVD(J) = 2 orthogonal modesx

SVD(J)

=x2

x1

Negative values = phase

U1U1* ε (0,.9) U2U2

* ε (-.1,.4)

Page 29: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

A Simple Example: Prior Knowledge

A B

C

3 modes hitting unknown structure

(A=.4, B=.5, C=.7)

+

SNMF(J)

SVD(J) = 2 orthogonal modes

=

x

x2

x1

Symmetric NMF: Assume no phase change- 3 modes>0, more info about structure

+ +

U1U1* ε (0,.47) U2U2

* ε (0,.6) U3U3* ε (0,.6)

Negative

Page 30: Partially Coherent Ambiguity Functions for Depth-Variant Point Spread Function Design Roarke Horstmeyer 1 Se Baek Oh 2 Otkrist Gupta 1 Ramesh Raskar 1

Keeping More than One Mode

= λ1 + λ2 +…+ λ3

Several Modes: Partially Coherent

x1

x2

x1

x2

-More accurate estimate with n>1 mutual intensity modes-Jest= Σ Ji <-> AFest = Σ AFi = Σ L(Ji)1,2 (L=linear transformation)-If Jest more accurate, then AFest more accurate

J

Multiple Modes can be: a. Multiplexed over time (Desantis, Zheng)b. Could also multiplex over space and/or angle

1M. Bastiaans, JOSA 3(8) 1986, 2Lohmann and Rhodes, Appl. Opt. 17, 1978