Face recognition using separate layers of the RGB image

In many cases face recognition of still images is performed with greyscale images. These images are actually converted from a color image to greyscale before the analysis takes place. A consequence of such a conversion is obviously loss of information, which could influence the performance of the face recognition system. It would be interesting to see if using one of the three color layers of the RGB image could give better recognition performance compared to the greyscale converted image. We conducted two experiments and the results indeed support this idea. We found that the red layer of the RGB image gives the best recognition performance, especially in the cases where an extra light source is used to light up (part of) the face of the participants in the experiments. In the case that the participants were facing the camera we saw the Equal Error Rate drop from 3.3% for the greyscale images to 1.8% for the red layer of the RGB images in our initial experiment.