Statistical Modeling and Learning of Visual Patterns
Lecturer: Song-Chun Zhu |
| |
Topic 1: Introduction to modeling, learning, and knowledge representation. |
The pursuit of Statistical models, Image space, Perceptual space, Three families of models,
Basic concepts in Information theory, Maximum likelihood estimation, Statistical
observations in the space of natural images, Scaling issues.
|
Topic 2: Learning with flat descriptive models
|
Maximum entropy principle, Markov random fields, Graphical models, Ising/Potts
models, FRAME theory, Minimax entropy principle, Julesz ensemble, Ensemble
equivalence theorem
Related computing issues: Relaxation labeling, Line drawing interpreatation,
Gibbs sampler,Swendsen-Wang cuts
|
Topic 3: Learning with hierarchic generative models |
FRAME theory, Sparse coding, Wavelets, Matching pursuit, Image pyramids;
Stochastic context free grammar, Learning and computing issues with SCFG
Related computing issues: Heuristic search algorithms,
Maintaining the OPEN-CLOSED lists,
Parsing algorithms in language with CFG,
Metropolis-Hastings, Reversible jumps.
|
Topic 4: Primal sketch: Mixing structures and Textures |
Integrated models, SCFG+bi-gram;
Image primitives, “lego” lands; Iimplicit and explicit manifolds in image spaces
|
Topic 5: Information scaling, Perceptual scale space |
Scale invariants, Information scaling laws;
Regimes in the image spaces, Perceptual transition of statistical regimes;
Perceptual scale space theory.
|
Topic 6: Stochastic context sensitive graph grammar |
Visual vocabulary, Configurations, Parsing graphs, And-Or Graphs;
Learning with the And-Or Graph--- MLE and pursuits
Related computing issues: Bottom-up / Top-down inference with grammar
maintaining lists of particles. |
|
|