Pattern and Object Recognition
| Course Number | EE 458b/892b |
| Course Title | Pattern and Object Recognition |
| Instructors | Peter N. Belhumeur |
| Dunham Labs, Rm. 506 | |
| Phone: 432-4249 | |
| belhumeur@ledoux.engyale.edu | |
| David J. Kriegman | |
| Dunham Labs, Rm. 502 | |
| Phone: 432-4091 | |
| kriegman@yale.edu | |
| Teaching Assistant | TBA |
| Schedule | TTh 2:30-3.45 at Becton 508 |
| Text | Pattern Classification and Scene Analysis, Duda and Hart |
Grading:
| Problem Sets | 20% |
| Paper Presentations | 20% |
| Literature Review/Proposal | 20% |
| Projects | 40% |
The course will develop mathematical techniques for both pattern and object recognition, with lectures equally split between both topics. The first half of the course will be devoted to statistical techniques for pattern recognition. An extended introduction will be given to topics in pattern classification, including Bayes decision theory, parameter estimation and unsupervised learning, bias vs. variance, nonparametric techniques, and linear discriminant functions. Neural networks will be introduced and discussed within the context of more classical pattern recognition schemes. In addition, more complex pattern recognition models -- i.e. Markov chains and Hidden Markov Models (HMM) -- for computer vision and speech recognition will be developed. The second half of the course will be devoted to geometric based approaches to object recognition. Geometric and algorithmic issues in recognizing three dimensional objects from an arbitrary viewpoint with partial occlusion will be considered. Building on affine and projective geometry, geometric invariants can be derived for certain object classes and used for indexing into a model database. For more general classes of 3D objects, model-based recognition algorithms rely on a 3D object representation. For objects with curved surfaces, differential geometry and topology provide the tools for relating image features to object models. Specific recognition strategies including interpretation trees, alignment, aspect graphs, geometric hashing, generalized Hough transforms, and pose clustering, and their complexity will be considered. Multi-part representations, object classes, context, and functionality will also be explored.