A novel gradient synthesis-based illumination formalization method for face recognition under varying lighting conditions

Face recognition remains a difficult area even though it has been researched for several decades. One of the challenges in automatic face recognition is to achieve illumination invariance. Performance of nearly all the current recognition system degrades due to variations by illumination. This paper proposes a novel illumination formalization method: Gradient Transform Synthesis. This method is based on applying different image transforms to the distinct parts of a facial image according to the lighting direction and illumination intensity. Experiments show this method is helpful and efficient for face recognition while dealing with variations caused by illumination, particularly under extremely lighting conditions.