Part VI: High-level Vision: Probabilistic and Inferential Methods
Finding Templates using Classifiers
Classifiers, including using loss, methods for building classifiers, an example of a plug-in classifier, a non-parametric classifier using nearest neighbours and estimating and improving performance
Building classifiers from class histograms, including finding skin pixels using a classifier and face finding assuming independent template responses
Feature selection, including principal component analysis, identifying individuals with PCA and canonical variates
Neural networks, including minimizing the error, when to stop training, finding faces with neural networks and convolutional neural networks
The support vector machine, including linearly separable datasets and finding pedestrians with SVM's
Appendices: Backpropagation, SVMS for non-linearly separable datasets and SVMS with non-linear kernels
Finding objects by voting on relations between templates, including voting and a simple generative model, probabilistic models for voting, voting on relations and voting and 3D objects
Relational reasoning using probabilistic models and search, including correspondence and search, and finding faces
Using classifiers to prune search, including identifying acceptable assemblies using projected classifiers and finding people and horses
Hidden markov models, including formal descriptions and algorithms
Application: HMM's and sign language understanding, including language models
Simple relations between object and image, including relations for curved surfaces and class-based grouping
Primitives, templates and generalized inference, including generalized cylinders as volumetric primitives, ribbons and representations built with ribbons
