Abstract
M.Phil. (Electrical and Electronic Engineering)
Identi_cation photos of human faces are required to meet a standard set by the ISO/ICAO organizations, for quality or for visibility since identi_cation photos are a type of biometric used mainly for access control. Some of the quality constraints that need to be met are the uniformity of the background, that there should not be more than one face on the photograph, that no shadows appear behind the person or on their face etc. The _rst step to be taken before developing an automatic system which will be able to detect whether all of these constraints have been compliant with is to _rstly know the structure of an image, that is the system needs to know where the person in the image is situated (hair, face/skin, torso and background). This can potentially be solved by segmentation of the di_erent regions of the person in the image. In this dissertation a region based segmen- tation using several methods available in image segmentation litera- ture is proposed to segment the person from the background. The results obtained showed the capability to segment facial images irre- spective of background illumination levels, texture and variation in size. These results are displayed in the dissertation of how the al- gorithm performed. It was observed that the algorithm performed better on uniform background facial images compared to other back- ground types. The other background types also performed well, but not as good as facial images with uniform background.