Finding image features using deformable templates and detailed prior statistical knowledge
Much work in image processing is essentially a bottom-up approach. An image is analysed by first applying simple image filters. The output from these filters is then grouped to recognise features and, subsequently, recognise objects. The work described in this report details essentially top-down met...
| Main Author: | |
|---|---|
| Published: |
University of Aberdeen
1992
|
| Subjects: | |
| Online Access: | http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.332373 |
| Summary: | Much work in image processing is essentially a bottom-up approach. An image is analysed by first applying simple image filters. The output from these filters is then grouped to recognise features and, subsequently, recognise objects. The work described in this report details essentially top-down methods. The algorithms also use simple image filters but search for known objects by using geometric models of the object outlines. The algorithms use statistical knowledge about the variation of the objects being searched for in order to guide the search to only feasible regions. The main techniques used are various deformable template algorithms where optimisations are achieved by random sampling and simulated annealing to avoid non-global extrema. The particular application here is for locating facial features including head outlines, where the results give key locations on the face and allow approximate geometric representations of the features to be reconstructed. |
|---|