Research Interests
Hidden Markov Model Learning Theory
Investigating new methods for learning HMM parameters more efficiently, yielding better classification performance. We have shown that similar classification performance can be achieved with a tenth of the training examples. Work is continuing to find a good theoretical framework for further advances.
Smart Camera Technology
By combining CMOS sensors, with FPGA signal processing and IEEE 1394 (Firewire) high-speed interconnects, we are in the process of building smart cameras dedicated to the purpose of computer vision rather than TV and video recording. The cameras are software reconfigurable for many applications including high-speed multi-lane numberplate recognition.
Face Recognition
We have found a new method for extracting illumination invariant features from principal components analysis (�eigenfaces�). This algorithm should have a major impact on the robustness of face recognition technologies. Other aspects of face recognition and affective computing (determining emotions from facial expressions) are current research themes within the group.
Image Segmentation with Geodesic Level Sets
Recent work with the CSIRO CMS has produced optimal segmentations of convex and concave objects. These results are better than those available through commercial tools. Earlier work in medical image segmentation within the CRC for Sensor Signal and Information Processing resulted in an international patent in 1999 and $1.5 million in contract research and licence agreements.