Chapter 3 describes a detection algorithm that utilizes biologically motivated low-level operations to find feature points on the face. A hierarchical, coarse-to-fine search is described for localizing the face. A method for the coarse detection of possible face regions in the image is outlined. We then propose a technique for the estimation of facial contour. Subsequently, we describe techniques for finding the eyes, the mouth and the nose in the face. We then describe the localization of the iris in the eye if it is visible in the image.
Chapter 4 describes the normalization procedure used to generate high-quality mug-shot images. It also discusses the linear transformations used to recognize them and to optimize their localization. We define the 3D deformable model used for normalization and the pose estimation computations that generate frontal, mug-shot images. The illumination correction algorithm is also presented. We then describe the Karhunen-Loeve decomposition and its use to measure the ``faceness'' of an image. We then describe how we improve the localization of detected feature points by maximizing the ``faceness'' value. Finally, we describe the use of the Karhunen-Loeve decomposition to recognize the identity of the subject in the image.
The details of the implementation and the output of the algorithm are covered in Chapter 5. Performance analysis and sensitivity analysis is performed to test the algorithm.
Finally, Chapter 6 concludes the thesis with a summary of the work, its contributions and the direction of future research and improvements.