Siddhartha Kasivajhula
PSYCH 221/ EE 362 (Winter 2007-2008)
Final Project: A Probabilistic Model of the Visual System

A complete MATLAB implementation of the model as described in the paper. This
includes:
1. generating an initial random scene,
2. evolving it through the Hidden Markov Model
3. Simulating an observation (as defined in the model)
4. Representing a prior knowledge of objects on the part of the observer
   - objects are represented as vectors of feature weights (see paper)
5. interpreting the observation as defined in the model -- by assigning weights
   to hypothesis objects, and identifying the object as the max. weight hypothesis


~~~~Files~~~~

GenerateInitialScene.m
GenerateScene.m
HMM_scene.m
MakeObservation.m
NoisyOrBN_interpret.m
ObserveScene.m
RepresentObserverKnowledge.m
TestModel.m

~~~~~~~~~~~~~

**************************************************************************
The file TestModel.m tests the entire implementation of the model, so you
probably want to run this file.
**************************************************************************

The file 'ObserveScene.m' generates a scene and makes an observation using the functions 'GenerateScene' and 'MakeObservation', so this is the file you'll want to run to test functionality of Representation and Observation.


Note: These files are only for illustration of how this model can be represented in MATLAB, and by no means represent the best way to do so. In particular, observations are made here using a "Roulette Wheel" sampling algorithm, which is not a very good method for this application, since the observations returned using this method can be completely different from the actual features contained in the visual scene if the complete feature space is much larger than the number of features contained in the scene (this isn't really important. I just want to communicate that this current implementation is inadequate for real modeling purposes. Shouldn't be hard to come up with a good implementation, though).

Also, some elements of this implementation are simplified from the model
described in the paper, and where such simplifications have been made, they
have been noted in the code.