6.881 | Spring 2005 | Graduate

Representation and Modeling for Image Analysis


Course Meeting Times

Lectures: 2 sessions / week, 1.5 hour / session


  • Subspace (Manifold) learning Theory: PCA Applications: Eigen faces, Active Shape & Active Appearance Models. Additional topics: kernel PCA, LLE
  • Boundary Detection Theory: Calculus of variations Applications: Mumford-Shah functional, snakes, level sets
  • EM Theory: EM algorithm Applications: segmentation, tracking
  • Graph algorithms Theory: Graph cut algorithms Applications: segmentation, stereo
  • Clustering Theory: hierarchical, k-means, spectral Applications: grouping in images
  • Graphical Models Theory: MRFs, inference in graphical models Applications: regularization, part/layer models
  • Shape descriptors Shape context, SIFT Medial axis, skeletons
  • Transformations and their manipulation Theory: diffeomorphisms, splines Applications: shape representation, registration
  • Information Theoretic Methods Theory: entropy and mutual information Application: alignment, segmentation
  • Classification Theory: nearest neighbor, perceptron, Fisher Linear Discriminant, SVMs, Ada Boosting Applications: object detection/recognition


This reading seminar aims to build up the mathematical background necessary to read papers and follow modern research in computer vision, as well as to improve communication skills, such as presenting research work, reviewing papers, surveying a field.

Everyone participating in the class must read the papers and come to class with questions on the assigned paper and on how it relates to other methods that attempt to solve the same problem.

Everyone will also be expected to present one or two papers during the semester and lead the discussion after the presentation.

Grading Policy

Method/Paper Presentations 40%
Participation in the Discussions 20%
Final Paper and Presentation (Project or Analysis Paper) 30%
Paper Review 10%