Project I: Feature selection for text categorization
Text categorization is a classification problem that aims at text contents to predefined classes. This problem is quite important as the amount of text data is increasing exponentially [1].
Several steps needs to be considered when implementing text categorization including, word stemming, stop-word removal, feature extraction, feature selection, classification and post processing. This project aims at investigating the effect of feature selection on text categorization.
[1] T. Giorgino, An Introduction to Text Classification, University of Pavia, 2004.
Project II: A Brain-Computer Interface System: Controlling computers by thought
Attempts to use thought-power alone to control computers have traditionally used electroencephalograph (EEG) signals, to measure the brain's electrical activity. The subject wears a skull-cap studded with electrodes. Because the skull muffles much of the neuronal chatter, the EEG records only large-scale activity, when large groups of neurons fire together. The major challenge of the brain-computer interface problem is how to pick up individual thoughts [1].
The two main issues that need to be performed in this project are (i) multi-channel processing, and (ii) classifying brain activity during a mental task.
Independent component analysis (ICA), which is a powerful technique that has the capability to separate mixtures of independent sources, will be used to process the multi-channel EEG signals. In order to handle the non-stationary nature of EEG signals, features will be extracted using discrete wavelet transform (DWT). Classification of mental tasks will be performed using artificial neural networks (ANN).
Regerences:
[1] Toby Howard, Controlling computers by thought, Personal Computer World magazine, February 1999.
Project III: Object tracking using deformable models
Real-time object tracking is critical task in many computer vision applications such as surveillance, perceptual user interfaces, augmented reality, smart rooms, object-based video compression, and driver assistance [1].
Object tracking in image sequences is usually based on the segmentation of moving objects in front of a stationary background. Tracking concentrates on algorithms to follow the target in a sequence of images [2].
The goal of this project is to build deformable models to represent and track objects. For this task we need to characterize objects by profile information, grey-level (or colour) distribution, and motion.
References:
[1] D. Comaniciu, , V. Ramesh and P. Meer, Kernel-Based Object Tracking, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 25, no. 5, may 2003
[2] A. Koschan, S. Kang, J. Paik, B. Abidi, and M. Abidi, "Color active shape models for tracking non-rigid objects," Pattern Recognition Letters, Special Issue on Color Image Processing, Vol. 24, pp. 1751-1765, July 2003.
Project IV: Audio-visual integration for improved speech recognition
Humans utilize multiple types of sensory information to communicate with each other. This is essential for more accurate, robust, natural and friendly interaction. These types of information are also required by machines to realize robust and friendly interfaces [1]. Accordingly, audio-visual processing has recently attracted much attention to overcome certain problems of audio-only speech processing. Difficulties due to background noise and multiple speakers are significantly reduced by the additional information provided by extra visual features [2].
This project is divided into three parts. The first part involves acoustic speech processing, which includes:
The second part is related to visual speech processing that includes
The third part is concerned with synchronisation and integration of the audio and visual parameters.
Note: Two students are needed to implement this project.
Reference:
[1] S. Nakamura, Statistical multimodal integration for audio-visual speech processing, IEEE Transactions on neural networks, vol 13, no. 4, July 2002.
[2] EK Patterson, S. Gurbuz, Z. Tufekci, and JN Gowdy, CUAVE: A New audio-visual database for multimodal human-computer interface research, Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, Orlando, May 2002.