This paper aims to investigate four various recognition approaches currently used in the technology of facial recognition at different stages of the operation. The four different recognition techniques, namely, normalised cross-correlation, gradient image, relative gradient image feature and eigenface methods were compared. The two main literatures cited were papers by Turk and Pentland [1] and Wei and Lai [2], based on Eigenface and Relative Image Gradient approaches respectively. Experimental results revealed that this face detection method is capable of detecting faces in different backgrounds, different illumination conditions, multiple faces in a single video frame, and head of varying sizes in the image. The face recognition results revealed that although the recognition techniques are incapable of face recognition that is invariant to severe illumination variations, yaw angle variations of head, and facial expressions variations, the eigenface method is proven to be quite tolerant with slight variations of these variables and with its superior speed performance. It was deemed the most appropriate for implementation for the complete system of face recognition in comparison with the others.

Keywords: Eigenface, Principal Component Analysis, Relative Image Gradient, normalised cross-correlation, gradient image background subtraction.