Department of Computer Science, Unit Catalogue 2008/09 |
CM20220 Fundamentals of pattern analysis |
| Credits: 6 |
| Level: Intermediate |
| Semester: 2 |
| Assessment: CW25EX75 |
| Requisites: |
| Before taking this unit you must take CM10197 |
|
Aims: To provide mathematical foundations underpinning concepts of digital signal processing (DSP), probabilistic methods and search/optimization algorithms demanded by Pattern Recognition, Computer Vision and Sound.
Learning Outcomes: On completion of this unit students will be able to: 1. understand and apply common signal processing operations, understanding their effect in terms of spatial and frequency domain. 2. understand and apply basic probabilistic methods to infer meaning from images. 3. understand and apply appropriate search strategies and optimization technique to solve Computer Vision problems. Skills: Problem Solving (T/F, A), Application of Number (T/F, A). Content: FUNDAMENTALS OF DIGITAL SIGNAL PROCESSING (DSP) * Frequency space - Continuous and Discrete Fourier transform (1D and 2D). - Discrete signal and image representation. Aliasing. - Taylor expansion * Digital signal processing (1D and 2D) - Convolution theorem - Low pass filters and antialiasing. Overview of high pass filters. - Image transformation, interpolation PROBABILITY AND PATTERN CLASSIFICATION * Means, standard deviation and variance. * Bayes theorem. Bayesian Inference. Priors. * Principal Component Analysis. Eigenmodels. * Classification - Concept of a feature space - Distance metrics - Supervised vs. Unsupervised learning. * Gaussian Mixture Models SEARCH AND OPTIMIZATION * Deterministic methods - Least squares - Best first, Gradient descent * Stochastic methods - k-means - Simulated annealing - Evolutionary search e.g. genetic algorithms |