PhD Proposals
Supervisor: Dr M Spann
High compression head and shoulders image coding
This project will investigate high compression image coding algorithms for head and shoulders colour images. Applications include encoding photographs on credit cards and video conferencing application. The approach is likely to be a combination of model-based and wavelet/VQ-based. Image parameters such as skin texture type, hair colour, face contour information etc. could be transmitted very efficiently and differences from the model encoded with traditional wavelet based algorithms.
"Event" detection within a video surveillance system
In most video surveillance systems, the vast majority of the time is spent surveying uninteresting scenes such as empty car parks or constant traffic flows. When an ‘event’ occurs such as an intruder entering the scene, it would be useful to detect these events and index the video frames for possible later retrieval. The project will investigate algorithms for determining a ‘normal’ pattern of video frames and an ‘abnormal’ pattern of frames. In particular, a generic system could be envisaged whereby the system could ‘learn’ normality in a given (possibly moving) scene and detect deviations from this normality.
There is a need for high performance medical image compression algorithms given the heavy use of 3D scanners and the need to archive the huge amounts of data obtained. Brain and body scan data contain a large amount of textured information corresponding to soft tissue, bone, organs etc. This project will consider the use of a hybrid coding method whereby texture is modeled in the wavelet domain using a small number of parameters in the higher frequency sub-bands. If this procedure is coupled with a wavelet packet representation of the image data, then high performance is achievable. Also, given knowledge of the imaging modality, texture parameters for models of different texture types could be learnt offline and applied to appropriate regions.
