Abstract:
Overall, humans are the most accurate face recognition systems. People recognize faces as part of social interactions, at a distance, in still and video imagery, and under a wide variety of poses, expressions, and illuminations. These conditions are challenging for computers. In this talk I will make the case that within five years it is possible for systems to perform general-purpose face recognition better than an average person. While algorithm development is the main challenge, there are other challenges that need to be addressed. One key obstacle is knowing when the performance of general-purpose algorithms is better than humans, which requires establishing a measurable goal line. With a goal line and an appropriate set of video and still images, one can construct a challenge problem for developing algorithms that are superior to humans for general face recognition. Based on previous experience, once a challenge problem is made available to the research community, rapid progress in made.