An Illumination Insensitive Normalization Approach to Face Recognition Using Locality Sensitive Discriminant Analysis

In this paper, a novel algorithm for face recognition is proposed in case of the images having illumination artifacts. First homomorphic filtering is done on the input face images to achieve partial illumination insensitivity. The fraction of the value of the image gradient to the original image intensity is evaluated to get an illumination independent normalized image. Here, gradient-domain is preferred since it explicitly accounts for the relationship between neighboring pixel points in the image. Then, Locality Sensitive Discriminant Analysis (LSDA) is applied to analyze the class relationship between data points. The proposed method performs very well, even if the number of training images is not sufficient. The experimental results on the extended Yale B database show that a significant improvement has been achieved in the recognition rate by making them illumination independent.