Comp 254, Image Processing and Analysis Spring 2002, Prof.
Guido Gerig
Syllabus
Goal: To present fundamentals for methods of
processing and analysis of images using a computer. Statistical pattern recognition,
multiscale shape and geometry, local to global grouping, model-based segmentation,
shape representation and analysis.
Approximate Outline and Nr. of Weeks
- Statistical pattern recognition for pixel processing: 3.0 w
- Classification, clustering, pixel-based image segmentation
- Karhunen Loeve transform, PCA, Fisher linear discriminant
- Mixture model approach
- Extraction of local primitives: 7.0 w
- Elementary geometry of surfaces: contours and ridges (Haralick)
- Image Segmentation as geometric analysis
- Concept of edge detection and operator design (Canny)
- Properties of Gaussian filtering, linear scale space
- Gaussian derivatives, invariants, multi-scale representation (scale space primal sketch, local
jets)
- Multi-scale detection of ridges and valleys by scale linking
- Linear scale space as a PDE
- Nonlinear diffusion (variable conductance diffusion,
nonlinear PDE)
- Grouping of local primities to global shapes: 6.0w
- Object recognition by Hough Transform
- Object recognition by invariant moments
- Object segmentation by deformable
contours (snakes)
- Explicit snake parametrization
- Implicit snake parametrization, geodesic snakes
- Object representation by Medial Axis (concept,extraction)
- Object segmentation by deformable medial model (m-rep)
- Contour description by Fourier harmonics
- Object segmentation by deformable parametrized contour model
- (**) Image Registration and Warping
- Linear and nonlinear transformations in
2D and 3D
- Approaches to calculate transformation
- Manual landmarking and TPS interpolation
- Feature space methods: Cross correlation ratio / Mutual Information
criterion
- Towards high dimensional warping
(**) Subject to availability of
time
Organization
| Teaching: |
Guido Gerig (gerig@cs.unc.edu, office SN219, phone 962 1919) |
| TA: |
Mathieu Ruffin (ruffin@cs.unc.edu) |
| Class time and place: |
9:30am-10:45am Tuesday and Thursday, SN011 |
| Prerequisites: |
1) Image processing, display and vision contents of COMP 235: i.e., sampling
and basics in digital signal processing, Fourier analysis, convolution
and filtering, the Gaussian and its derivatives
2) Calculus, linear algebra, probability theory
|
Material
| Handouts |
Copies of relevant book chapters and publications |
| Recommended books |
- Bernd Jahne, Digital Image Processing, Springer, 1997 (General Image
Processing Book, very comprehensive, HIGHLY RECOMMENDED)
- Milan Sonka et al., Image Processing,Analysis, and Machine Vision, PWS Publishing, 1998 (General Image Processing Book, Comprehensive, RECOMMENDED)
- R.O. Duda, P.E. Hart, D.G. Stork: Pattern Classification, Wiley Interscience,
2000 (Statistical Pattern Recognition)
|
Assignments
Assignments Theory and Methods, solutions on paper (50%) / Programming
work using Matlab, or other tools of choice (50%)
Grading (subject to modifications)
5 assignments (with 2 weeks time) 60% Final exam 35% Class participation
5%
FINAL EXAM DATE Sat May 4, 2002, 9 pm (maybe a take-home exam)
Class participation
Regular attendance is necessary to follow the subject as the teaching will
complete and enhance the material found in the handouts. Please inform the
TA (ruffin@cs.unc.edu) early enough if you can't attend a lecture due to adverse
circumstances.
Guido Gerig, January 10, 2002
This document was generated on 9 January 2002 using the texi2html
translator version 1.54.