Face recognition (FR) systems can perform well under uncontrolled illumination, but there is no general and robust technique with total immunity to all conditions. Hence, investigating it still is an open research area. In this work, we investigate methods that can be applied to illumination changes and, with this, it is proposed an illumination invariant FR approach that employs unlighting photometric-based techniques together with a Local Binary Patterns (LBP) texture descriptor. To conduct the experiments, we used the leave-one-out accuracy estimation model using the well known Yale B face dataset with different degrees of illumination. The results show that our methodology is robust since it achieves 100% FR rate in the most challenging subsets.