| This dissertation presents an integrated approach to modeling and detecting semi-rigid bodies in intensity images by the use of deformable templates. This integrated approach is used within the human face recognition domain in order to demonstrate its benefits. A thorough discussion of the current approach to deformable templates is presented and its limitations are highlighted. Four new energy measures are created, which overcome these limitations, and in order to demonstrate their performance four new face organ templates (eyebrow, nose, face, and chin) are create which would be difficult with the current approach.; Deformable templates require to be initialized close to their target object, in order to provide an automated system, a new novel algorithm which estimates face organ location from intensity images is also presented. The deformable templates are then minimized using a modified Greedy algorithm. Once the templates have found all the face organs, geometric measures of the face are taken. These measures are then used within a Bayesian framework in order to recognize the face. |
