Ch.10 Image Understanding

464p

10.3 Point distribution models

http://en.wikipedia.org/wiki/Point_distribution_model

Autostitch

1. Aligning the training data
the transformations to reduce (in a least-squares sense) the difference between an aligned shape and a 'mean' shape

2. Deriving the model
The eigen-vectors of then 2N*2N covariance matrix S will provides a basis, meaning that we can represent any vector x as a linear combination of the 2N different p_i.

http://en.wikipedia.org/wiki/Principle_Component_Analysis
the components of a basis vactor indicate how much variation is exhibited with respect to each of the eigenvectors

-> dimensinal compression of the representation

3. Fitting models to data

http://en.wikipedia.org/wiki/Eigenvalue,_eigenvector_and_eigenspace

http://en.wikipedia.org/wiki/Active_shape_model

Tim Cootes <An Introduction to Active Shape Models>

asm_overview.pdf

- overview paper about ASMs and AAMs (a useful introduction) 

asm_aam_overview[1].pdf

4. Extensions



475p

10.4 Active Appearance Models

1. Modeling image appearance


http://en.wikipedia.org/wiki/Hidden_Markov_model