CPSC 505 Home Page

This is World Wide Web home page for UBC course CPSC 505 Image Understanding I: Image Analysis, as taught by Jim Little, 2000.

Lectures: Tue. and Thu. 1300-1430, MCML 158

The average 505 face 2000


Course Description

The current UBC calendar description for the course is:

CPSC 505 (3) Image Understanding I: Image Analysis -- Image formation constraints and the processing of digital images in order to extract information about the world being imaged. Computational methods for image analysis. [3--0; 0--0].


Course Outline

We will study image formation constraints and techniques for analyzing digital images to determine information about the world being imaged. This course provides the basic tools for later research presented in CPSC 525. Understanding digital images requires a combination of physics, electronics, mathematics, and computational theory. During this course we develop the necessary tools for analysis of images and for understanding what is possible to determine from an image. We will cover topics from image formation (optics), image structures (geometry and computational theory), binocular stereo and motion (mathematics -- analysis and geometry), the relation of computational vision to human vision (psychology), and finally the computational techniques for analyzing images and recovering scene properties (signal processing and computer science).

Tentative Lecture Schedule

  1. Introduction:
    • basic definitions
    • image understanding: image analysis and scene analysis
    • image formation: projective geometry - perspective and orthogonal
    • digital images: quantization, tesselation and noise
    • the adequacy of digital images: bandlimited signals, sampling theorems
  2. Digital image processing:
    • linear shift-invariant operations: convolution
    • template matching: global templates, local templates, matched filtering
    • transform domain approaches: the Fourier transform, Gabor functions
    • linear systems theory as an analysis tool
  3. Edge detection:
    • Physical causes of edges
    • Marr--Hildreth theory of edge detection
    • Canny edge detection (with Deriche extensions)
    • Torre--Poggio regularization
    • Blake--Zisserman GNC algorithm
  4. Shape from shading and photometric stereo:
    • modeling image formation: geometry, radiometry and the image irradiance equation
    • intermediate representations: visible surfaces, the 2.5-D sketch, intrinsic images
    • orientation-based representations of 3-D shape
  5. Optical flow and the 2-D motion field:
    • gradient (i.e., differential) approaches
    • derivation of motion invariants
    • local versus global constraints
    • multiple light source optical flow
  6. Selective introduction to other current research in image analysis:
    • other vision modules: binocular stereo, colour, texture
    • vision problems formulated as minimization problems
    • more mathematical tools: morphology, Markov random fields, mean field theory
    • multiscale analysis: the wavelet transform, pyramids
  7. Applications to robotics:
    • vision systems for recognition, localization and inspection
    • motion tracking
    • navigation
  8. Summary and suggestions for further work


Course Text

The recommended text for the course is "Robot Vision" by B.K.P. Horn. BUT you need not buy it. It will be on reserve and all materials for the course will be presented in lectures.
The errata for the text can be found in postscript form in Robot Vision errata.

Additional texts used in the course:


Lecture Notes

For further information, please refer to the lectures91 or lecture notes in postscript, reorganized Sep. 97, from the originals of 1991.

You can also get the individual sections of the 1997 reorganization of the notes:


Sample Final Exam

The December 96 final exam is at: December 96.

Several questions on the 96 final were not handled in the 00 version:
1a, 2c, 5; the rest you should be able to answer. Read them carefully.


Course Assignments

For information about the assignments, see 505 Assignments.


Images we use

TIFF file for image TIFF file for image
TIFF file for image TIFF file for image TIFF file for image TIFF file for image TIFF file for image TIFF file for image


Additional Material

For help on Projective Geometry, see the postscript version of the tutorial on Projective Geometry.

And Projective Geometry by Stan Birchfield.

For help on the Fourier Transform, see the tutorial on the 1D Fourier Transform.

Some material on radiometry (courtesy Alain Fournier and Paul Lalonde) radiometry.

There is a wealth of information about Computer Vision at the Computer Vision Home Page maintained at CMU: Vision Home Page.
You can find test images, and assorted interesting demos there.

The Perceptual Science Group at MIT has some demos at Demos.


Illusions

Some cool visual illusions: visual illusions

More cool illusions: visual illusions

More cool illusions: visual illusions


Content-Based Image Retrieval

Berkeley
Cornell


Materials for Vision Courses Elsewhere

Forsyth and Ponce (Berkeley and Illinois) have a draft version of their text for computer vision. I expect this will be the new standard text: Forsyth and Ponce David Heeger at Stanford has put his lecture notes for "Vision and Image Processing" on the Web. Highly recommended.

You can get the course materials for a course in Vision at Cornell at Cornell.

Course notes from another institution (mentioned in class):Notes.

Industrial Vision

Cameras and Equipment

Interesting Links