information-theoretic signal and image processingjao/talks/csttalks/jaostalk6.pdf · 2004-06-10 ·...
TRANSCRIPT
Information-Theoretic Signal and Image Processing
Professor Joseph A. O’Sullivan Electrical and Systems Engineering
Associate Director, Center for Security Technologies
Science and Technology Intellectual Thrusts
• Information-Theoretic Signal and Image Processing:Snyder, Byrnes, Fuhrmann, Ghosh, Isidori,
Mukai, O’Sullivan, Preza, Wickerhauser
• Recognition Theory and Systems:O’Sullivan, Byrnes, Chamberlain, Grimm,
Indeck, Martin, Pless, Smith, Wickerhauser
Outline: Signal and Image Processing
• Fundamentals of Imaging Science• Application: Recognition System Design and Analysis
– Open problem: fundamental performance bounds as a function of system complexity
– Systems integration issues for biometrics• Application: X-Ray Imaging for Luggage Inspection
– Open problem: implement efficiently, combine recognition– Systems integration for combined imaging, threat detection
• Application: Spectral Analysis– Open problem: performance for spectral unmixing– Systems integration for rapid analysis, interpretation
Technology Expertise and Enablersdevices, design, analysis, computation, electronics, systems ...
• Cheap, even disposable sensors• Inexpensive mass data storage• Fundamental advances in imaging science• Reconfigurable, fast electronics • Affordable, rapid computational processing• Economical, pervasive networking, including
wireless
Imaging Science at Washington University
• Leadership in Medical Imaging— Development of First PET Machines— Brain Mapping— Deformable Anatomy
• Astronomical Imaging, Hubble Telescope• Center for Imaging Science, 1995-2000
— Fundamentals of Imaging Science, Recognition Theory— Performance Bounds for ATR— ATR System Design and Analysis
CST: applying analogous methodology to security technologies
Fundamental Approach
SceneModel
PropagationModel
SensorModel
AlgorithmDevelopment
AbstractionBarrier
PerformanceAnalysis
Physics of Scenesand Spectra
PropagationPhysics
Physics of Sensorsand FPAs
Signal and ImageProcessors
• Biometrics-based recognition including face,fingerprint, voice, retinal, DNA, etc.
• Physical signature recognition including magnetic signatures
• Fast database searching and recognition system implementation
• Data mining, intelligence extraction, situational awareness
Target Target ClassifierClassifier ââ=T72=T72
Recognition Theory and Systems
Recognition Theory and Systems
• System Design – Data models system outline– Training data, physics model parameters– Implementation constraints on architecture,
computational platform, time– Performance goals and constraints
• Range of Applications– Object recognition– Biometric recognition– Biochemical contaminants
Example: Radar ATR Theory and Performance
• Publicly available SAR images from MSTAR program; over 15,000 images in data set
• Compare performance and database complexity for a variety of data models
2S12S1 T62T62 BTR 60BTR 60 D7D7 ZIL 131ZIL 131 ZSU 23/4ZSU 23/4
• Derive statistical models from data, physics, constraintspR|Θ,a(r|θ,a) - conditional data modelpΘ|a(θ|a) – prior probability on orientationp(a) – prior probability on target class
• Recognition: select most likely target• Evaluate performance over test set• Repeat for different models and complexity constraints
Target Target ClassifierClassifier
ââ=T72=T72Given a SAR image r, determine a corresponding target class â∈A and pose θ
Target Recognition ProblemStatistical Likelihood Approach
Recognition Performance
•Model correctness: validity of statistical likelihood•Model segmentation: parts of data modeled
•Approximation complexity: number of poses
BMP2 Variance Image at 6 Sizes
Fine model of a T72 tank (1/7 relative scale),72 templates totaling 1.1M floats
Coarse model of aT62 tank, 1 template with 16K floats
Example Results•240 variations per model families•10 model families (6 reported)
Performance—Complexity Analysis
• Open Problems– Given finite size database, finite
representation of data, determine fundamental limit on performance
– Computational limit
• Systems Integration– Design of biometric systems– Dynamic resource allocation
Target Target ClassifierClassifier
ââ=T72=T72
CT Imaging: Luggage Inspection
• Mandate that every checked bag be screened as of Jan. 1, 2003 (by the Transportation Security Administration (TSA))
• Goals: accurate images, good display, automatic detection of potential threats, speed
• WU Basis: medical CT imaging, nondestructive testing
CT Imaging
• Physics-based models• Novel algorithm development to
mitigate effects of metal• Medical imaging for prostheses,
brachytherapy applicators, dental implants
• Industrial imaging nondestructive testing
CT Imaging
• Challenges – efficient implementation– unknown objects in scene, unknown
orientations, shapes– spectral modeling
• Security Problem: Screening of Luggage– accurate artifact-free images– automatic threat detection– speed of processing
Spectral Analysis
• Security Applications– Detection of biological and chemical contaminants
• Security of food and water supply, mail, currency• Emergency room detection, pattern detection
– Remote spectral sensing• Agricultural classification
– Mass spectrometry• Systems Integration Issues
– Rapid analysis, extraction of signatures– Rapid search of databases– Coupling to sensors of interest
Spectral Unmixing
• Component Spectra– Database– Extracted from data
• Quantify Variability ofSpectra
• Quantify Relative Concentrations inNew Data
Conclusions
• Fundamental engineering and scientific approach to signal and image processing
• Critical link in many security systems• Tight coupling between
• Information-Theoretic Signal and Image Processing
• Recognition Theory and Systems
• Unique expertise at Washington University