A study on the effective approach to illumination-invariant face recognition based on a single image

In this paper, the methods for single image-based face recognition under varying lighting are reviewed. Meanwhile, some representative methods as well as their combinations are evaluated by experiments, and the underlying principle of the experimental results is investigated. According to our investigation, it is almost impossible to attain a satisfied face recognition result by using only one facial descriptor/representation especially under drastically varying illuminations. However, the "two-step" framework, including an illumination preprocessing and an illumination-insensitive facial features extraction, could be an effective approach to addressing this problem. We further study what are the appropriate illumination preprocessing and feature extraction for this framework.