Automatic face detection and recognition has been a difficult problem in the field of computer vision for several years. Although humans perform the task in an effortless manner, the underlying computations within the human visual system are of tremendous complexity. The seemingly trivial task of finding and recognizing faces is the result of millions of years of evolution and we are far from fully understanding how the brain performs it. Furthermore, the ability to find faces visually in a scene and recognize them is critical for humans in their everyday activities. Consequently, the automation of this task would be useful for many applications including security, surveillance, gaze-based control, affective computing, speech recognition assistance, video compression and animation. However, to date, no complete solution has been proposed that allows the automatic recognition of faces in real (un-contrived) images. We wish to develop a vision system which would permit automatic machine-based face detection and recognition in uncontrolled environments. This thesis describes the theoretical foundations of such a system, its implementation and its evaluation.
This introduction begins with an outline of the main issues and constraints that need to be addressed in face recognition. Subsequently, a survey is presented which outlines the face recognition research that has been performed to-date and the strengths and weaknesses of a variety of machine-based systems. We then describe our proposed approach for the face recognition problem. Finally, an overview of the structure of this thesis is presented.