CS 580  
Fall 2003
Dr. Duric					Due:  12/2/2003


			Final Project


You are to select an area of interest and do a final project in that
area.  A list of topics to chose from is given below.  If you wish,
this may be a team project (2-3 people).  In that case more will be
expected than from 1-person projects.

Schedule:
   By Oct. 1: obtain agreement with me on the project.
   Dec. 2:  hand in project report.


List of topics:

1. Map coloring 

Explore map coloring in depth. Implement heuristics listed in the
textbook and  experiment with different maps. See how random state
ordering influences the efficiency of your heuristics. See your
teacher for additional maps. You should compare running times and
number of steps needed for various heuristics.

2. Railroad puzzles

Use iterative deepening search to solve the railroad puzzles given at
http://www.cs.gmu.edu/~zduric/cs580/Railroad/index.html
What is the branching factor for those puzzles?

3. Constraint satisfaction

Develop a program that uses min-conflict heuristics to solve n-queens
problem for at least 100,000 queens. See the textbook for details. 

4. Machine Learning

Develop a program which learns how to do things better over time.  A
simple example of this is a program which learns to play the game of
Nim.  See the file 3-pile-nim.txt for more details.

5. Vision

a) Use a publicly available implementation of an edge detector
   (e.g. Canny's edge detector - options: see your instructors,
   Matlab, Computer Vision Home page) to detect edges in a coridor
   image. For random pairs of edge pixels check if they belong to the
   same image line.  This can be done by generating a digital line
   through the pixels and checking how many edge pixels belong to that
   line (Bresenham algorithm from computer graphics can be used to
   generate a line). How many random pairs are needed to find all
   interesting lines?

b) Use publicly available implementation of an edge detector
   (e.g. Canny's edge detector - options: see your instructors,
   Matlab, Computer Vision Home page) to detect edges in a face
   image. Use matlab to label eyelid pixels and approximate the eyelid
   edges using the spline toolbox. If you are given all edges in the
   eye area is it possible to check random combinations of points to
   find eyelids?

c) Consider the outline of a walking human (see your teacher for
   example images). Approximate the boundary by as few straight lines
   as possible to obtain a polygonal shape - hand pick boundary points
   for this approximation. Can you think of a way to find body parts
   and label them as legs, arms, torso, head, etc. You can decompose
   the body into smaller parts/polygons. Your teacher has images. You
   can use matlab for this project.

d) Consider a 2-second sequence of a walking human. Use matlab
   to hand label the corresponding stick figures/skeletons. If you
   view the sequence of skeletons can you recognize the motion? See
   your instructor for images.

6. Neural Networks

Using one of the neural network packages available on the net, develop
a small project using neural networks to recognize simple objects such
as hand-drawn letters or numbers.

7. Genetic Algorithms

Using one of the genetic algorithms packages available on the net,
develop a small project using genetic algorithms as a heuristic
search mechanism for solving an NP-hard problem like TSP or SAT.

8. 2x2x2 Rubik's cube

Consider a simplified Rubik's cube (2x2x2 - the original cube is
3x3x3). Formulate the representation and the operators for this
problem. Find what is the depth of the search space for this
problem. Choose a search algorithm and solve if it for a variety of
start and goal states. Note: this may not be as trivial as it sounds.

The final report should include a detailed hardcopy report on what you
did, how you did it, and how it turned out. You should acknowledge all
help that you received including all information obtained from various
web sources. It should include program source and sample output.  In
addition, source code and executable files should be made available.