CS 534
Computer Vision


Spring 2003
W 3:00 - 4:30 PM, Hill 254
F 2:30 - 4:00 PM, CoRE A
Basics
 Organization
 Syllabus
 Homework
 Projects
 Tests
 Resources

Basics

Computer Vision / CS534 is an introductory graduate-level course.  It is meant to provide a  broad coverage of modern computer vision problems, techniques, and methodologies.  The course will cover fundamental vision issues as well as the latest and greatest (and not so great!) results from the field.

Instructor

Vladimir Pavlovic
Office: 312 CoRE
Email: vladimir@cs.rutgers.edu
Web: www.cs.rutgers.edu/~vladimir
Phone: 732-445-2654

Office hours

Mondays, 3-4.

Mailing list

cs534-spring03@rams.rutgers.edu
List archive.

TA / Grader

TBA

Textbook

"Computer Vision: A Modern Approach"
By David Forsyth and Jean Ponce
Prentice Hall 2002
Amazon BN Buy.com

Prerequisites

Knowledge of linear algebra and stochastic processes is essential.  If you don't know how to compute a matrix inverse please consider taking a linear algebra course first.  The same  goes for stochastic processes.  Ask yourself if you know what a Gaussian distribution or a maximum likelihood estimates are.

Familiarity with basic image processing techniques will help you increase the slope of your learning curves.  You should also be familiar with Matlab and C or C++, as most of the assignments will  require one of the two (most likely Matlab).  

Organization

Grading

Homework
 15%
Midterm
 20%
Project
 30%
Final
 30%
Participation
 5%

Midterm and final will not include any programming.  Homework will. Project should heavily rely on programming unless you are proficient in theory or hardware.  Participation is mandatory!

Syllabus

Week
 		Topic
1
 Introduction
Camera models; Lenses
2
 Radiometry;
Local shading models; Photometric stereo
3
 Color
4
 Linear filters; Edges
5
 Texture
Multi-view geometry
6
 Stereopsis
7
 Project abstracts due, March 5, 2003
Midterm: March 7, 2003
8
 Segmentation
9
 Basic model fitting (lines, curves)
10
 Tracking
11
 Model-based vision
12
 Template matching using classification
13
 Deformable models
14
 Structure from Motion & optical flow
15
 Project presentations
May 2, 2:00pm in CoRE A.


Homework

Homework will be assigned bi-weekly.  Most homework assignments will include programming problems.  You are encouraged to consult your fellow students on the homework problems, but please turn in your original solutions.  This particularly holds for the programming problems.

I expect all homeworks to be turned in as assigned.  Late submissions will not be tolerated.

Assignment
 		Problem Set
 		Due Date
 		Solutions
1
 PS1
 1/31/2003
 SS1
2
 PS2
Images
 2/26/2003
 SS2
3
 PS3
VisTex
 3/5/2003
 SS3
4
 PS4
 4/2/2003
 SS4
5
 PS5
Images
 4/17/2003
 SS5
6
 PS6
 4/30/2003
 SS6

Projects

Projects count!  Do you have the Lego Mindstorms vision set?  How about some handheld-device (e.g., Sharp Zaurus) vision?  Or a distributed activity recognition project? Of course, you can pick a number of interesting topics.

Possible project topics:

1. Motion clustering using hidden Markov models.
Implement a hidden Markov model-based method for clustering of motion trajectories, similar to Smyth et al. See also Pavlovic et al. for human motion modeling. Extend to recognize human motion styles and types, similar to Tanenbaum and Freeman.

2. Bayesian color constancy.
Implement and experiment with the Bayesian color constancy algorithm proposed by Brainard and Freeman et al. Consider possible extensions to their approach.

3. Motion segmentation using Bayesian Networks.
Implement the Transformed Markov Model framework proposed by Jojic et al. for layered object segmentation.

4. Real-time Vision on handheld devices.
Develop a vision application that will run in real-time on a handheld device (IPaq, Sharp Zaurus). Color constancy, specular removal, object detection, ...

5. Loopy Stereo.
In a recent paper by Shum et. al., a existing Bayesian inference algorithm known as loopy propagation was applied to the problem of stereo matching. (Ask me for a reference to this work.)

6. Object Recognition.
Develop a program for detecting coins in images using Hough transform. Extend your program to recognize dollar bills as well as coins (using the same edge-based technique). Extend your program further to reason about possible occlusions of coins by dollar bills. (See textbook for references.)

7. Texture Synthesis.
Implement the texture synthesis procedure described in the paper by Heeger and Bergen. Validate your algorithm on the textures from their paper, and then explore the limitations of the algorithm. Try using this framework for texture matching/discrimination.

Project presentations:

Project presentations will be on May 2 at 2:00pm in CoRE A. Each pressntation should be 20 minutes long (15 presentation + 5 questions). Please cover the following in your presentation: motivation, prior approaches, your solution, implementation, results/experiments, and conclusions.

Project reports:

Project reports are to be turned in before the final exam. Please, do not make them longer than 20 pages.

Tests

March 7, 2003. Midterm will cover lectures 1-8. Problems and solutions.

May 9, 2003. 2:30pm, Hill 254. Final will cover lectures 9-end. Problems and solutions.

Scores

Click here to see homework/test/project scores.

Resources

Useful matrix and Gaussian identities by Sam Roweis.

Matlab

Documentation from Mathworks.
Tutorial from MTU.
Tutorials from University of Colorado.
Image processing tutorial from UCSD.
Probabilistic Modeling Toolkit.