Title: Terahertz Imaging with Compressed Sensing
1Terahertz Imaging with Compressed Sensing
Wai Lam Chan
Department of Electrical and Computer
Engineering Rice University, Houston, Texas, USA
December 17, 2007
2Terahertz (THz) Research Group at Rice
- Mittleman Group
- (http//www.ece.rice.edu/daniel)
THz Near-field microscopy (Zhan, Astley)
THz waveguides (Mendis, Mbonye, Diebel, Wang)
THz Photonic Crystal structures (Prasad, Jian)
THz emission spectroscopy (Laib, Zhan)
THz Imaging (Chan, Pearce)
3T-rays and Imaging
4What Are T-Rays?
T-Rays
X-Rays
Radio Waves
Hz
Visible Light
Microwaves
Gamma Rays
5Imaging Throughout History
Daguerreotype (1839)
X-rays (1895)
T-rays (1995)
http//inventors.about.com/library/inventors/bldag
uerreotype.htm
http//inventors.about.com/library/inventors/blxra
y.htm
B. B. Hu and M. C. Nuss, Opt. Lett., 20, 1716,
1995
6Why Can T-Rays Help?
Subpicosecond pulses
Linear Phase
Over 1 THz in Bandwidth
T-Rays Provide
Benefits to Imaging
- Travel-time / Depth Information
- High depth resolution
- High spatial resolution
- Measurement of E(t)
- Subpicosecond pulses
- Submillimeter Wavelengths
7Material Responses to T-rays
Plastics
Transparent
Metal
Highly Reflective
Water
Strongly Absorbing
8Promising Applications of T-Rays
Medical Imaging
(Kawase, Optics Photonics News, October 2004)
Diseased Tissue
Concealed Weapon
Wallace, V. P., et. al. Faraday Discuss. 126, 255
- 263 (2004).
Security
Safety
Zandonella, C. Nature 424, 721722 (2003).
(Karpowicz, et al., Appl. Phys. Lett. vol. 86,
054105 (2005))
Space Shuttle Foam
9THz Time-domain Imaging
THz Transmitter
THz Receiver
Object
10THz Time-domain Imaging
- Pixel-by-pixel scanning
- Limitations acquisition time vs. resolution
- Faster imaging method
Just take fewer samples!
11Compressed Sensing (CS)
Candes et al, Donoho
12Why CS works Sparsity
- Many signals can be compressed in some
representation/basis (Fourier, wavelets, )
13High-speed THz Imaging with Compressed Sensing
(CS)
- Take fewer ( ) measurements
- Reconstruct via nonlinear processing
(optimization)
(Donoho, IEEE Trans. on Information Theory,
52(4), pp. 1289 - 1306, April 2006)
14Compressed Sensing (CS) Theory
- Signal is -sparse
- Few linear projections
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
R
measurements
sparsesignal (image)
Measurement matrix
information rate
15Compressed Sensing (CS) Theory
- Signal is -sparse
- Few linear projections
- Random measurements will work!
-
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
R
measurements
sparsesignal (image)
Measurement matrix (e.g., random)
information rate
16Random can be
Random 0/1 (Bernoulli)
1
2
M
Random 2-D Fourier
1
2
M
and many others
17CS Signal Recovery
- Reconstruction/decoding given(ill-posed
inverse problem) find
sparsesignal
measurements
nonzeroentries
18CS Signal Recovery
- Reconstruction/decoding given(ill-posed
inverse problem) find - L2 fast, wrong
19CS Signal Recovery
- Reconstruction/decoding given(ill-posed
inverse problem) find - L2 fast, wrong
- L0 correct, slow only MK1 measurements
required to perfectly reconstruct
K-sparse signal Bresler Rice
number ofnonzeroentries
20CS Signal Recovery
- Reconstruction/decoding given(ill-posed
inverse problem) find - L2 fast, wrong
- L0 correct, slow
- L1 correct, mild oversampling Candes et
al, Donoho
linear program
21CS in Action Part I CS-THz Fourier Imaging
22THz Fourier Imaging Setup
object mask
THz transmitter (fiber-coupled PC antenna)
metal aperture
THz receiver
R
6cm
6cm
6cm
6cm
automated translation stage
23THz Fourier Imaging Setup
Fourier plane
object mask
N Fourier samples
THz transmitter
R
6cm
6cm
6cm
24Random 2-D Fourier
Measurement matrix
25THz Fourier Imaging Setup
automated translation stage
THz receiver
object mask R (3.5cm x 3.5cm)
polyethlene lens
26Fourier Imaging Results
6.4 cm
4.5 cm
R
6.4 cm
4.5 cm
Resolution 1.125 mm
Inverse Fourier Transform Reconstruction
(zoomed-in)
Fourier Transform of object (Magnitude)
27Imaging Results with CS
4.5 cm
4.5 cm
CS Reconstruction (500 measurements)
CS Reconstruction (1000 measurements)
Inverse FT Reconstruction (4096 measurements)
28Imaging Using the Fourier Magnitude
object mask
metal aperture
THz receiver
THz transmitter
R
6cm
6cm
variable object position
translation stage
29Reconstruction with Phase Retrieval (PR)
- Reconstruct signal from only the magnitude of its
Fourier transform - Iterative algorithm based on prior knowledge of
signal - real-valued
- positivity
- finite support
- Hybrid Input-Output (HIO) algorithm
- Compressive Phase Retrieval (CPR)
(Fienup, Appl. Optics., 21(15), pp. 2758 - 2769,
August 1982)
(Moravec et al.)
30Imaging Results with Compressive Phase Retrieval
(CPR)
6 cm
6.4 cm
R
6 cm
6.4 cm
Resolution 1.875 mm
Fourier Transform of object (Magnitude-only)
CPR Reconstruction (4096 measurements)
31Compressed Sensing Phase Retrieval (CSPR) Results
- Modified CPR algorithm with CS
6.4 cm
6 cm
6 cm
6.4 cm
Fourier Transform of object (Magnitude-only)
CPR Reconstruction (4096 measurements)
CSPR Reconstruction (1000 measurements)
32CS in Action Part I CSPR Imaging System
- THz Fourier imaging with compressed sensing (CS)
and phase retrieval (PR) - Improved acquisition speed
- Processing time
- Potential for
- Flaw or impurity detection
- Imaging with CW source (e.g., QCL)
33CS in ActionPart II Single-Pixel THz Camera
34Imaging with a Single-Pixel detector?
- Continuous-Wave (CW) THz imaging
- with a detector array
- Real-time imaging
(Lee A W M, et al., Appl. Phys. Lett. vol. 89,
141125 (2006))
35Single-Pixel Camera (Visible Region)
R
DSP
imagereconstruction
DMD
DMD
Random pattern on DMD array
(Baraniuk, Kelly, et al. Proc. of Computational
Imaging IV at SPIE Electronic Imaging, Jan 2006)
36Random 0/1 Bernoulli
.001010.
Measurement matrix
37Random patterns for CS-THz imaging
- Random patterns on printed-circuit boards (PCBs)
38THz Single-Pixel Camera Setup
Random pattern on PCBs
object mask
THz transmitter (fiber-coupled PC antenna)
THz receiver
R
7cm
42cm
6cm
39THz Single-Pixel Camera Imaging Result
CS resconstruction (200 measurements)
CS resconstruction (400 measurements)
Object mask
40THz Single-Pixel Camera Imaging Result
CS resconstruction (400 measurements)
CS resconstruction (200 measurements)
41CS in ActionPart II Single-Pixel THz camera
- First single-pixel THz imaging system with no
raster scanning - Potential for
- Low cost (simple hardware)
- near video-rate acquisition
- Faster acquisition
- film negatives (wheels/sprockets)
- more advanced THz modulation techniques
42Conclusions
- Terahertz imaging with Compressed Sensing
- Acquire fewer samples high-speed image
acquisition - THz Fourier imaging with CSPR
- Single-pixel THz camera
- Ongoing research
- THz camera with higher speed and resolution
- Imaging phase with CS
- CS-THz tomography
- Imaging with multiple THz sensors
43- Mittleman Group (http//www.ece.rice.edu/daniel)
- Contact info William Chan (wailam_at_rice.edu)
Acknowledgement Dr. Daniel Mittleman Dr. Richard
Baraniuk Dr. Kevin Kelly Matthew
Moravec Dharmpal Takhar Kriti Charan
dsp.rice.edu/cs
44T-Ray System
THz Transmitter
Femtosecond Pulse
Substrate Lens
GaAs Substrate
Picometrix T-Ray Instrumentation System
Picometrix T-Ray Transmitter Module
Femtosecond Pulse
DC Bias
45T-Ray System
Sample
THz Transmitter
THz Receiver
Optical Fiber
T-Ray Control Box with Scanning Delay Line
Fiber Coupled Femtosecond Laser System
46Summary of T-Rays
- Broad fractional bandwidth
- Direct measurement of E(t)
- Short wavelengths (good depth resolution)
- Unique material responses
47Sampling
- Signal is -sparse
- Samples
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
R
sparsesignal
measurements
nonzeroentries