Algorithm evolution for face recognition: what makes a picture difficult

One of the classic problems in computer vision is the face recognition problem. In general, this problem can take on a wide variety of forms, but the most common face recognition problem is "Who is this a picture of?" Evolutionary computation has, in the past, been applied indirectly to this problem through techniques like learning neural networks. This paper introduces a genetic programming style approach to learning algorithms that directly investigate face images and are coordinated into a face recognition system. Through a series of experiments, we show that evolved algorithms can accomplish the face recognition task. We also highlight several pitfalls and misconceptions surrounding face recognition as a learning problem.