One problem with this approach is that it would be relatively easy to train classifiers that are looking for people who are attempting to avoid detection.
Wear highly reflective object like a diamond on the face, preferably on the forehead. It will project light to the camera and create random pattern on the face. What do you think?
It might be worthwhile to consider that detecting a face and finding the identity of a face are not the same. From the "Eigenface" article on wikipedia,<p>"The pictures constituting the training set should have been taken under the same lighting conditions, and must be normalized to have the eyes and mouths aligned across all images."<p>I would be curious to see how well face identification works with various milder forms of camouflage.
While this can evade detection using full-facial classifiers, it's not difficult to detect individual facial features. If you can detect the eyes, nose, and mouth, and have them in a region in an orientation matching a face, you could assume that it's probably a face. While it is less efficient it's still a possibility.