In the course a series of advanced image analysis techniques will be presented through lectures and computer exercises.
Frequently, in image analysis problems we have well structured prior information about the visual phenomenon that we are observing. In the course we will formulate the inclusion of such prior information in the analysis by use of the Bayesian paradigm. The Bayesian paradigm combines prior information and observations in a statistical setting.
We will review the Bayesian paradigm and utilise it for a series of purposes (e.g. segmentation, regularisation, restoration).
The prior models considered in the course include Markov Random Field image models for modelling texture and deformable template models for modelling shape.
Also, we will consider the analysis of multivariate image data. Here the purpose is extraction of relevant information (data mining).
Furthermore, we consider the application of greyscale mathematical morphology for extracting structures (e.g. blobs) from images.
Geostatistics is the statictics of spatial dispersed data. We will shortly review the method og kriging (linear interpolation between randomly sampled data).
Finally, we will consider the problem of infering (3D) structure and scene-camera motion from a video stream. |