In case of incremental inputs to an online face recognition with illumination invariant face samples which maximize the class-separation criterion but also incorporates the asymmetrical property of training data distributions. In this paper we alleviate this problem with an incremental learning algorithm to effectively adjust a boosted strong classifier with domain-partitioning weak hypotheses to online samples, which adopts a novel approach to efficient estimation of training losses received from offline samples. An illumination invariant face representation is obtained by extracting local binary pattern (LBP) features NIR images. The Ada-boost procedure is used to learn a powerful face recognition engine based on the invariant representation. We use Incremental linear discriminant analysis (ILDA), in case of sparse function for active near infrared (NIR) imaging system that is able to produce face images of good condition regardless of visible lights in the environment accuracy by changes in environmental illumination The experiments show convincing results of our incremental method on challenging face detection in extreme illuminations.