ECE-537 Digital Image Processing

1
ECE-537Digital Image Processing

• Prof. K. J. Hintz
• http//cpe.gmu.edu/khintz
• http//cpe.gmu.edu/

2
Course Objectives

• Develop an Understanding of Basic Image
  Processing Techniques Through Lecture, Study, and
  Exercises
• Develop a Fluency With a Digital Image Processing
  (DIP) Software Package, e.g., Cantata, Through
  Experience
• Implement an Independent DIP Project Which
  Demonstrates Your Ability to Integrate the
  Mathematical Theory With the Practical Issues

3
Milestone Chart
4
Khoros Image Processing Software

• Composer
• Toolboxes
• Cantata (our primary interest)
• Workspaces for interactive DIP programming
• Glyph-based scripting language
• Workspaces are in ASCII text format

5
Image Processing Software

• Khoros Will Be Installed Soon on iris.gmu.edu
• Class List Will Be Given to Computer
  Administrator to Put in Group for Access to
  Software
• Students can get combination to ST-II room ___
  to have access at non-class times
• Khoros can be run effectively on a sufficiently
  powerful PC running the Linux operating system
• Windows version requires Xwin32 X

6
Khoros Availability

• Students wishing to use Khoros at home can
  download the student version from the Khoros web
  site at no charge
• Build it yourself, straightforward
• http//www.khoral.com/khoros/kp2001_student/
• Binaries available from Amazon for 30

7
Linux Operating System

• PC based, x86
• The more RAM the better
• Linux can be downloaded free
• http//www.ibiblio.org/pub/Linux/distributions/
• Various distributions available on CDROM (40)
  with various levels of support

8
Linux Components Needed to Install/Use Khoros

• gzip
• ANSI-C compiler, e.g., gcc
• X-Windows (X11R4, R5, or R6) Software Development
  Environment, e.g., XFree86
• Athena, Motif, or OLIT widget set
• Lex YaCC (or bison flex)
• Optional
• perl
• groff (needs C compiler)
• Fortran Compiler or f2c (matrix geometry TB)

9
Useful Free Linux Software

• GIMP
• GNU Image Manipulation Program
• XV
• Display images
• Image format conversion
• Color manipulation
• Some image processing operators
• Smooth
• Dither
• Enlarge
• Oil paint

10
Digital Image

• N-Dimensional Data, Not Necessarily Visual Image
• Passive Sensors
• Infrared
• Ultraviolet
• Visual
• Multi-spectral, hyper-spectral
• State of point
• Active Sensors
• Imaging radars
• Synthetic aperture radar (SAR)
• X-ray

11
Passive Sensor Data

• Infrared
• Single-band/Dual-band
• 3-5 micron and/or 8-12 micron wavelength
• Greyscale image
• Ultraviolet
• Visual
• Greyscale
• HSI

12
Passive Sensor Data

• Multi-spectral, hyper-spectral
• Each pixel is n-dimensional vector
• Subdivides sensor bandwidth with narrow-band
  filters
• State of point

13
IR Images of Clouds

Near-IR, 3-5 ?
Far-IR, 8-12 ?
14
UV Image of Foraminifera
15
Visual Image, BW vs. Color

Same spatial resolution
20 kByte
29 kByte
16
Digital Image

• Computed Data
• Function representation
• Visualization of complex functions, e.g.,
  Mandelbrot set
• Virtual reality
• Fabric draping

17
Active Sensor Data

• Imaging Radars
• 95 GHz gt ? 1/8 inch
• Frequency chosen based on atmospheric absorption
  bands
• (Inverse) Synthetic Aperture Radar ((I)SAR)
• Multiple measurements from long baseline of
  moving sensor
• ISAR uses movement of target
• X-ray

18
SAR Image of Airport

Synthetic Aperture Radar
19
X-Ray MRI
Khoros
20
Digital Image

• Quantized
• Amplitude
• Intensity
• Absorption
• Numerical value (magnitude)
• Color or pseudo-color
• Signal strength
• Spatially (Khoros notation)
• Height in rows
• Width in columns

21
Synchronous Amplitude Quantization

22
Spatial Quantization

23
Processing vs. Analysis

• Digital Image Processing (DIP)
• manipulation of images by computer
• Texture and edges
• Segmentation
• Enhancement
• Mensuration
• Movement
• Feature extraction

24
Analog Image Processing

• Analog Image Processing
• Optical SAR, Fourier transform
• Film processing
• Image summation
• Color, e.g., TV RGB

25
Processing vs. Analysis

• Image Analysis
• Pattern recognition
• Automatic control
• Content
• Operates on features or symbolic representation
• Optical character recognition (OCR)
• Handwriting
• Intelligence

26
Fractal Dimension of IR Clouds

27
Character Recognition
Extract Rotate Segment Classify Parseno
28
Image Spatial Manipulation
Matlab Example
29
Foreground/Background

Marr, Computer Vision
30
Foreground/Background
31
Digital Image Processing

• Digital Image
• ...a sampled, quantized function of two
  dimensions that has been generated by optical
  means, sampled in an equally spaced rectangular
  grid pattern, and quantized in equal intervals of
  amplitude.
• Processing
• ...starts with one image and produces a modified
  version of that image without interpreting it

32
Discrete vs. Continuous Approaches

• Image Consists of Discrete Points
• Each pixel is a unique, independent entity
• Process points rather than entities in image
• e.g., calibration of column of individual
  detectors used to generate image
• Assumes no underlying continuous process

33
Discrete vs. Continuous Approaches

• Continuous Approach
• Underlying process is continuous, e.g., optical
  image of real object
• Processing mathematics are continuous with a
  discrete approximation, e.g, Fourier Transform
  and DFT

34
Digital Processing of Images

• Completely Accurate Only when Original Data is
  Quantized and Output is Quantized
• Usually Done for Convenience
• Digital Image is Approximation of Real,
  Continuous Image
• Digital Operations are Approximations of
  Continuous Mathematical Processes

35
Processing Artifacts

• Approximate Operations on Approximate Images can
  Yield Sampling Effects
• Moiré pattern (continuous also)
• Ragged edges
• Gibbs phenomenon

36
Gibbs Phenomenon

• Insufficient Bandwidth to Completely Represent
  Edge
• Simulated by
• Discrete Fourier transform
• Discard high frequency spectral lines
• Inverse DFT
• Results in Ringing Near Edges

Ziemer, Tranter, Fannin, Signals and Systems
Continuous and Discrete, Macmillan, 1989.
37
Square-in-Square Cantata Workspace

38
Gibbs Phenomenon Workspace

39
Expansion of S-in-S Glyph

40
Inset Glyph Menu Pane

41
PSD of S-in-S

42
Interactive Thresholding

43
Gibbs Phenomenon

44
Unified DIP Approach

• Characterize Effect of DIP from Knowledge of
  Continuous Forms Behavior
• A/D --gt Process --gt D/A Without Loss or
  Significant Degradation of Content of Interest
• Sampling and DIP Artifacts
• Predict their occurrence
• Recognize them when they occur
• Eliminate or minimize their deleterious effects

45
DIP is Multidisciplinary

• Preparation of Object to be Imaged
• Stain used for tissue (biology)
• Illumination (photography)
• X-ray energy and energy spectra (physics)
• Selection of amplitude and spatial resolution
  (communications)

46
DIP is Multidisciplinary

• Digitizing the Image
• Equipment selection (CCD, near-IR, far-IR, x-ray,
  etc.)
• Signal-to-noise ratio (communication theory)
• Full dynamic range of resulting image and bias
  errors (A/D converters, amplifiers, electronics)
• Purpose for images (end user)

47
Image-to-Image Consistency

• Amplitude calibration
• Grey-scale charts
• Visual scenes
• Stepped wedges
• X-ray absorption images
• Black-body
• Spectrum proportional to T4
• Hot body, cold body calibrate endpoints
• Multi-spectral

48
Image-to-Image Consistency

• Spatial calibration
• Fiducial marks
• INF Treaty X-Ray example
• Easily identifiable features
• Crosses on roadways for satellite landmapping
• Known geometry of acquisition system and scene
• Structured light example

49
Image-to-Image Consistency

• Image-to-Image Consistency
• Temporal imaging
• Intelligence analysis, airplanes in/out of
  airport
• Velocity estimation
• Database change
• Image registration
• Anomaly detection
• Known clutter removal

50
Image-to-Image Consistency

• Clutter Removal (image next slide)
• Real signals, not noise
• Characteristics of clutter
• Statistical properties
• Shape
• Random
• Geometric
• Distinctive features

51
Detect Scalloped Lilly Leaves
52
DIP is Multidisciplinary

• Processing
• End user
• What does the user want as a result of the
  processing
• What techniques can be applied (mathematics)
• Real-time or batch
• Numerical analysis
• Data type conversions
• Numerical processing errors
• Roundoff
• Truncation
• Order of computations

53
DIP is Multidisciplinary

• Display DIP Results
• Impedance matching
• Response of human eye and limitations
• physiology, pseudo-coloring
• Human signal processing capability
• psychology
• Still or motion
• weather map with fronts satellite

54
DIP Functional Requirements

• Sufficient Spatial Sampling Resolution to Have
  Multiple Pixels on the Feature of Interest
• Best Signal-to-Noise
• Fast, Large RAM/HD Computer
• High-Quality, Color Display

55
DIP Functional Requirements

• User-Friendly, Mathematically Correct and
  Verified Software
• Ease of Documentation
• Modularized to Ease Task of Explaining Processing
  to Others

56
Homework/Project Data

• Each student assigned an image
• Homeworks done plus principles of HW applied to
  assigned image
• Color result presented each week to class
• Semester project based on assigned images or
  negotiated alternative

57
Available Data Sets

• Foraminifera
• Separate forams from other objects
• Large images

UV
Visible
58
Automatic Foram Extraction
59
Available Data Sets

• Ground Penetrating Radar Data
• Volumetric processing
• Image processing downrange/crossrange
• High resolution images
• Clutter removal
• MTF

60
GPR DeMine program
61
Available Data Sets

• Handwritten Character Data Set
• High-resolution X-Ray images
• 9-track tape
• IR images of sea clutter
• Visual images
• Hyperspectral Satellite Imagery
• On Web