![]() However, a study performed from 2015 to 2017 with my colleagues at the University of Maryland Center for Automation Research, Alice O’Toole’s group at the University of Texas, Dallas, and Jonathon Phillips’s group at the National Institute of Standards and Technology found that while an AI-based algorithm performed better than human experts at recognizing faces, the most accurate results came when humans and computers worked together. I will consider the implications of the technology’s bias later in this chapter.Īrtificial intelligence is extremely powerful when matched to faces drawn from a large collection of images, including the billions of unique head shots taken over decades for driv er’s licenses, passports, and other forms of identification. Many now believe that all forms of AI could be biased, but in fact the bias has been extensively studied in this one application of deep learning. The findings attracted widespread media attention and have had profound implications for the entire field of artificial intelligence. ![]() In recent years, however, tests have shown that percentage to be lower for individuals with darker skin tones.Ī 2018 study by the Massachusetts Institute of Technology found that some of the commercial applications for facial recognition, when applied for gender classification, demonstrated gender and skin-tone biases. The matching accuracy of facial recognition algorithms today can exceed 95 percent-for certain categories of faces. Cameras and algorithms work better when faces and scenes are well lit, but less ideal conditions can be addressed by training the system on a larger set of faces so that it can recognize varied skin tones under different conditions. We’ve also found ways to use computational power to make processing faster and to deter “deepfake” attacks that try to trick algorithms into using false facial images. We’ve done so by accounting for variations in images of faces due to illumination, aging, and camera angles, as well as atmospheric distortions that come from images at a distance. My colleagues and I have spent the past three decades making these systems more accurate and efficient. These cameras capture still images and video sequences, and compare them against an existing database of faces, such as state drivers’ licenses or passport photos. The evolution from the bulky television cameras of the 1960s and 1970s to present-day SLRs and handheld smartphone cameras has contributed to the success of face recognition algorithms. As face recognition technologies improved, the quality of facial images also improved thanks to the advances made in high-resolution camera designs. Today, facial recognition is a software system that combines powerful digital cameras with deep-learning algorithms (computer programs that learn to process large amounts of data as the human brain would) that can match features from a person’s face to an existing database of images. It’s one of many biometric measurements that experts use to identify people including fingerprints, iris scans, speech recognition, and gait. My interest and work in facial recognition date back to the early 1990s, but computer vision and pattern recognition researchers and engineers have been developing the technology since the early 1970s. ![]() In this chapter, we’ll take a look at how the field of facial recognition and verification technology developed, how it works currently, where it needs adjustment, and what applications lie ahead. 2 Others note that, used properly, face recognition can speed identification of criminals keep buildings, property, and air planes safer and aid in the rescue of missing and exploited children. Advocacy groups such as the American Civil Liberties Union say that its use by law enforcement agencies is an invasion of an individual’s privacy, and they point to the technology’s biases against certain groups. 1 Use of a facial ID scanner to screen people entering a secure building may not prompt much concern, but many consider deploying similar technology along a public street or in a park as intrusive, perhaps even authoritarian. Half of US adults have their images stored in police face recognition databases, according to a 2016 report by Georgetown University’s Center on Privacy and Technology. As with almost any technology, the impact of this series of digital tools depends on the human decisions informing their programming and guiding their use in myriad environments, under differing circumstances, and affecting a variety of communities. ![]() CHAPTER 3 The Complexities and Contributions of Facial RecognitionįACIAL RECOGNITION IS THE APPLICATION of artificial intelligence that tends to sound the loudest alarms.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |