Separating useful from useless image variation for face recognition

For a general purpose face recognition system one of the largest challenge is to separate useful identity related from useless variations in the image data due to nuisance variables such as: orientation, lighting, expression, possible disguise. A recognition system is presented in which the effect of the secondary/nuisance variables is to a large degree accounted for before the matching process even begins. Greatly improved performance is shown on a large database of faces in 42 conditions.