**CS-644B: Pattern Recognition**

*"We
don't see things as they are. We see them as we are."* - Anais
Nin

**Detailed
Course Contents**

## Introduction
to Pattern Recognition via Character Recognition

- Transducers
- Preprocessing
- Feature extraction (feature-space
representation)
- Classification (decision regions)
- Grids (square, triangular,
hexagonal)
- Connectivity
- Contour tracing (square &
Moore neighborhood tracing)
- M.I.T. reading machine for
the blind
- Hysteresis smoothing (digital
filtering)
- Types of input to pattern
recognition programs

## Spatial Smoothing

- Regularization
- Logical smoothing (salt-and-pepper
noise)
- Local averaging
- Median filtering
- Polygonal approximation

## Spatial Differentiation

- Sobel operator
- Roberts cross operator
- Laplacian
- Unsharp masking

## Spatial Moments

- Moments of distributions
- Moments of area & perimeter
- Moments for feature extraction
- Moments for pre-processing
- Moments as predictors of discrimination
performance

## Medial Axis Transformations

- Distance between sets
- Medial Axis (prairie-fire
transformation)
- Skeletonization
- Hilditch's algorithm
- Rosenfeld's algorithm
- Minkowski metrics
- Distance transforms
- Skeleton clean-up via distance
transforms
- Medial axes via distance transforms

## Topological Feature
Extraction

- Convex hulls, concavities
and enclosures

## Processing Line Drawings

- Square, circular, and grid-intersect
quantization
- Probability of obtaining diagonal
elements
- Geometric probability (Bertrand's
paradox)
- Difference encoding &
chain correlation functions
- Minkowski metric quantization

## Detection of Structure
in Noisy Pictures and Dot Patterns

- Point-to-curve transformations
(Hough transform)
- Line and circle detection
- Hypothesis testing approach
- Maximum-entropy quantization
- Proximity graphs and perception
- Triangulations and Voronoi
diagrams
- The shape of a set of points
- Relative neighbourhood graphs
- Sphere-of-influence graphs
- Alpha hulls & Beta skeletons

## Neural Networks and
Bayesian Decision Theory

- Formal neurons, linear machines
& perceptrons
- Continuous and discrete measurements
- Minimum risk classification
- Minimum error classification
- Discriminant functions
- The multivariate Gaussian
probability density function
- Mahalanobis distance classifiers
- Parametric decision rules
- Independence and the discrete
case

## Independence of Measurements,
Redundancy, and Synergism

- Conditional and unconditional
independence
- Dependence and correlation
- The best k measurements are
not the k best
- Information theory and feature
evaluation criteria
- Feature selection methods

## Neural Networks and
Non-parametric Learning

- Perceptrons
- Non-parametric training of linear machines
- Error-correction procedures
- The fundamental learning theorem
- Multi-layer networks

## Estimation of Parameters
and Classifier Performance

- Properties of estimators
- Dimensionality and sample
size
- Estimation of the probability
of misclassification

## Nearest Neighbor Decision
Rules

- The k-nearest neighbor rule
- Efficient search methods for
nearest neighbors
- Decreasing space requirements
- Editing training sets
- Error bounds

## Using Contextual Information
in Pattern Recognition

- Markov methods
- Forward dynamic programming
and The Viterbi algorithm
- Combined bottom-up and top-down
algorithms

## Cluster Analysis and
Unsupervised Learning

- Decision-directed
learning
- Graph-theoretic methods
- Agglomerative and divisive
methods

*Teaching Activities*
*Homepage *