Recognizing Faces in Shades of Gray

Face recognition depends on relatively few distinguishing features, when compared with common facial features. This gives color information greater value to recognition and identification processes. However, dealing with grayscale facial images is a must in some cases, e.g. legacy images. In this paper, we investigate the effect of losing color information on face recognition. We propose a novel framework, which utilizes CNN-based colorization before a CNN classifier. The proposed framework is tested on LFW benchmark dataset. The evaluation results prove the success of the proposed framework in reducing the negative effect of dropping the color information on face recognition performance.