signal processing for computational imaging - minh n....
TRANSCRIPT
Signal ProcessingSignal Processingforfor
Computational ImagingComputational Imaging
Minh N. DoDepartment of Electrical and Computer Engineering
University of Illinois at Urbana-Champaign
1
20 Greatest Engineering Achievements20 Greatest Engineering Achievements of the 20th Century of the 20th Century
Selected by the U.S. National Academy of Engineeringwww.greatachievements.org
2
1. Electrification2. Automobile3. Airplane4. Water Supply and
Distribution5. Electronics6. Radio and Television7. Agricultural Mechanization8. Computers9. Telephone10.Air Conditioning and
Refrigeration
11.Highways12.Spacecraft13. Internet14. Imaging15.Household Appliances16.Health Technologies17.Petroleum and
Petrochemical Technologies18.Laser and Fiber Optics19.Nuclear Technologies20.High-performance Materials
20 Greatest Engineering Achievements20 Greatest Engineering Achievements of the 20th Century of the 20th Century
3
ImagingImaging
• In 1900…• Only simple microscopes, telescopes, and
black and white photographs.
• In 2000…• Color photographs, movies, and television• New sensors that greatly expand our “vision”
• see tiny atomic particles to vast galaxies• see inside the human body• track weather patterns, …
• The entire microelectronics revolution is basedupon imaging
4
ImageImage
FormationFormation
ImageImage
ProcessingProcessing
• Examples: optical imaging, radar imaging, computer tomography(CT), magnetic resonance imaging (MRI).
Imaging = Image Formation + ProcessingImaging = Image Formation + Processing
9
ImageImage
FormationFormation
ImageImage
ProcessingProcessing
Imaging = Image Formation + ProcessingImaging = Image Formation + Processing
Tradition scope ofTradition scope ofsignal processingsignal processing
11
Imaging is a Multidisciplinary FieldImaging is a Multidisciplinary FieldAnd Signal Processing is the GlueAnd Signal Processing is the Glue
MathematicsMathematics PhysicsPhysics
ApplicationApplication
domaindomain
ComputerComputer
ScienceScience
ComputationalComputational
ImagingImaging
12
An Application Domain: Health CareAn Application Domain: Health Care
• In 2005, the U.S. spent 16% of its GDP on health care.It is projected that this will reach 20% by 2015.• Starbucks: healthcare cost exceeds coffee.• General Motor: healthcare cost exceeds steel.
• A quarter of total Medicare budget is spent in the last year of life;40% of it is on the last 30 days.
Goal: Individualized treatments based on better medical imaging
• In developing countries: desire to have low-cost and effectivepopulation screening.
13
Hardware Approach toHardware Approach toImaging System CalibrationImaging System Calibration
• Require expensive hardware modification and calibration
14
Computational Approach to CalibrationComputational Approach to Calibration• Apply signal processing methods to form useful imagery from
imperfect signal measurements.• Use computational power to overcome physical limitations.
15
MRI with Field MRI with Field InhomogeneityInhomogeneity
• Field inhomogeneity is a key factor affecting MRI image reconstruction
datadata imageimagefield mapfield map
HarmonicHarmonicRetrieval !Retrieval !
16
Joint Estimation in MRI usingJoint Estimation in MRI usingHarmonic Retrieval MethodsHarmonic Retrieval Methods
Original image
and field mapStandard
reconstructionOur reconstruction
H. Nguyen et al., 2007H. Nguyen et al., 2007
17
Autofocus Autofocus Problem inProblem inSynthetic Aperture Radar (SAR)Synthetic Aperture Radar (SAR)
18
MCA: MCA: MultiChanel AutofocusMultiChanel Autofocus
wherewhere
R. Morrison et al., 2007R. Morrison et al., 2007
19
Beyond TelevisionBeyond Television…… Remote Reality Remote Reality
M. Maitre et al., 2007M. Maitre et al., 2007
21
Minimax Minimax Design of Hybrid Filter BanksDesign of Hybrid Filter Banks
H. Nguyen and MD, 2007H. Nguyen and MD, 2007
22
Image Processing:Image Processing:The Key is Efficient Image RepresentationThe Key is Efficient Image Representation• Goal: Parsimonious view of pictorial information
A A naturalnatural image image A A randomrandom image imageA A biologicalbiological image image
Natural images Natural images ““livelive”” in a very tiny bit in a very tiny bit
of the space of all possible of the space of all possible ““imagesimages””
23
OriginalOriginal domain: pixels domain: pixels WaveletWavelet domain: coefficients domain: coefficients
Image Representation via TransformsImage Representation via Transforms
25
Are We There?Are We There?
“…the 20 bits per second which, the psychologists assure us, thatthe human eye is capable of taking in…”
D. Gabor, 1959
One image Human eye (≈ 5 sec.) 100 bits
One imageJPEG-2000 (wavelets)
1’000’000 bits
26
What Have We Learnt from Sputnik?What Have We Learnt from Sputnik?
“The lesson of the past 50 years, however, is that the more humanitydiscovers about space, the rarer and more precious life on Earthseems.”
Economists, 27 Sep 2007