We have thus presented a method for normalizing images for 3D pose changes and illumination. This technique is employed with the Karhunen Loeve decomposition to statistically determine how face-like an image is. This helps guide a nose detection search which minimizes the ``DFFS'' value. Finally, the fully localized face is normalized and analyzed using the KL to match it to an image in the database. This final stage completes the algorithm proposed in Chapter 3 and generates recognition results.