CS 664 Computer Vision

CS 664 Computer Vision
Cornell University
Fall 2004

Lectures Tu Th 1:25-2:40, in 140 Bard.

This course is intended for graduate students and advanced undergraduates who have good programming skills (using C or C++), a good mathematics background, and a knowledge of algorithms. Students will be expected to pick up new mathematical techniques during the semester, as covered in lecture, and to relate the mathematical concepts from lecture to the programming assignments.

Here is an outline of the topics to be covered:

Optimization and vision: discrete versus continuous techniques, simulated annealing, graph cuts, belief propagation, level sets
Image restoration: local approaches (including smoothing, convolution, and robust estimation) and global approaches
Stereo matching: basic issues, local approaches, variable-windows, dynamic programming, Marr-Poggio, energy minimization
Image segmentation: region merging, graph-based problem formulations, modeling issues
Motion: Optical-flow constraint equation, Lucas-Kanade, layers, motion segmentation via expectation-maximization
Tracking: active contours (snakes), Hausdorff-based object tracking, CONDENSATION
Edge and feature detection: first-derivative (Canny-style) and second-derivative (Marr-Hildreth) edge detectors, Lowe's SIFT operator
Object recognition: Hausdorff distance and fraction, eigenspace methods, recognizing flexible objects
Texture: Gabor filters and textons, learning textures from examples, texture segmentation
Medical imaging: image formation (CT and MR), multi-modal registration, segmentation, contextual information
Statistical methods: ML/MAP estimation, expectation-maximization, particle filters, mutual information, density estimators (mean shift), learning (Viola/Jones face detector), robust statistics, Markov Random Fields, dimensionality reduction
3D vision: structure from motion, alignment, calibration

Course Requirements: There will be two programming assignments, a (short) oral paper presentation and a final project. Except for the oral paper presentation, each of these will require implementation, testing with image or video data, and a write-up. The programming assignments can be done in pairs, and will be due roughly 6 weeks and 12 weeks into the semester.

The programming assignments and final project will require you to use C/C++ on a Windows NT platform. (This class will not cover how to use a C development environment to complete the assignments.) In addition, there may be a series of (very short) 5-minute written quizzes in lecture, that will cover basic comprehension of the mathematical material.

Final Project: Since this is a 600-level course, the major requirement is that you do a short piece of original research. The research project must be done alone. As part of your project, you are expected to read a technical computer vision paper that is not covered in class, and to prepare a short (15-minute) oral presentation that gives an overview of the paper. The oral presentations will take place in the last week of classes; depending upon the schedule, some students may be asked to present to the entire class. You are expected to choose your paper, and to write a 1-paragraph project proposal, by November 1.