Facial recognition – a learning science (Part 1 of 2)
In general, consumers are excited to adopt new technologies in the latest upgrade of their favorite gadget or new tech designed to save time, decrease irritation, or increase the fun of daily life.
Facial recognition is a technology that was hardly mentioned outside of spy novels before the November 2017 release of the iPhoneX, with its Apple Face ID. But as quickly as the new phone’s popularity grew, interest in facial recognition spiked and press coverage was soon flooded with stories about the uses – and misuses – of the technology. What many do not realize, though, is that facial recognition technology has its roots in early Artificial Intelligence (AI) research, and that, like AI, the science of facial recognition is a growing, learning exercise that gets better as more data – primarily in the form of faces – is collected. In this 2-part post, we will look at how facial recognition came into being and how its development has been aided by increasingly large sample datasets and government-sponsored competition.
Initial research on facial recognition was done in the 1960s by Woodrow Wilson “Woody” Bledsoe – also one of the founders of the science of Artificial Intelligence – at his firm, Panoramic Research, Inc. (PRI) in Palo Alto, California.
Bledsoe was contracted by unnamed intelligence agency and worked with colleagues Helen Chan and Charles Bisson on the complex problem of using a computer to recognize a human face. The team reported limited success working with a book of mugshots as its facial “database”, but once Bledsoe left PRI, his work was picked up by Peter Hart at Stanford Research Institute, who had much greater success working with a database of 2000 photographs.
Even more progress was made in the late 90s by graduate students working at the University of Bochum, in Germany, and the University of Southern California, with funding from the US Army Research Laboratory and using a database of 14,126 images. At this point, commercial security use began at some banks and airports, as it was now able to make identifications with less-than-perfect face views and even through some recognition impediments, such as facial hair and sunglasses.
In Part 2 of this post, we will look at competitions and challenges that have been sponsored by the US government in an effort measure and compare the effectiveness and the capabilities of various commercial approaches to facial recognition.