Facial recognition system

Facial recognition is a specialized type of biometric technology. Biometrics refers to the measurement of a person's unique physiological characteristics to identify them. This system works by matching a human face from a digital image or a video frame against a stored database of faces. It is widely used for ID verification and authentication services. Because the process is contactless, it has seen massive global adoption. However, its accuracy is generally lower than other biometrics like iris recognition, fingerprint scanning, or voice recognition.

To identify a person, the computer must complete a specific four-step process. First, the system performs face detection to segment the face from the rest of the image background. Second, it uses alignment to adjust the image for the face's pose and size. It also accounts for photographic properties like grayscale or lighting levels. Third, the system performs facial feature extraction. During this stage, the computer pinpoints and measures specific parts like the eyes, nose, and mouth.

Finally, the system creates a feature vector, which is a mathematical representation of the face. This vector is then matched against a database of known faces to find a potential match.

The history of this technology began in the 1960s. Pioneers Woody Bledsoe, Helen Chan Wolf, and Charles Bisson worked to teach computers to recognize human faces. Their early project was known as "man-machine" because it required human assistance. A person would use a graphics tablet to mark coordinates for features like the pupil centers or the hairline. The computer then used these coordinates to calculate 20 individual distances, such as the width of the mouth. In 1970, Takeo Kanade demonstrated a system that could locate anatomical features like the chin without human help. Kanade later published the first detailed book on facial recognition in 1977.

In the 1990s, research shifted toward more automated methods. In 1993, DARPA and the Army Research Laboratory established the FERET program. This program evaluated different methods to see if they could work in real-life security and law enforcement settings. The FERET tests were successful enough to spawn three U.S. companies, including Vision Corporation and Miros Inc. During this decade, the technology moved into public service. For example, DMV offices in West Virginia and New Mexico were among the first to use these systems. They used facial recognition to prevent people from getting multiple driver's licenses using different names.

Mathematical models have changed how computers "see" faces over time. In the early 1990s, Matthew Turk and Alex Pentland developed Principal Component Analysis (PCA). This method is also called the Eigenface approach. It uses a linear model to represent a human face as a weighted combination of several "Eigenfaces." This greatly reduced the amount of data a computer had to process. Later, in 1997, researchers improved this using Linear Discriminant Analysis (LDA) to create "Fisherfaces." By the late 1990s, the Bochum system used Gabor filters to link facial features into a grid. This system was robust enough to identify people despite glasses, beards, or different hairstyles.

Modern applications have become even more integrated into daily life. In 2001, the Viola-Jones object detection framework made real-time face detection in video possible. By 2015, this algorithm was small enough to run on handheld devices and embedded systems. Today, the technology is used in many different ways. It is used in video surveillance, passenger screening at airports, and even in robotics.

Some governments even use it in specialized ways. For instance, Ukraine has used Clearview AI software to identify deceased Russian soldiers. This has been used as a form of psychological warfare to raise awareness of Russian activities.

Despite its utility, facial recognition is highly controversial. Many critics claim the technology violates privacy and can make incorrect identifications. There are also serious concerns regarding racial profiling and gender bias. The rise of synthetic media, such as deepfakes, has created new security worries. These issues have led to real-world consequences for major tech companies. In 2021, Meta Platforms shut down its Facebook facial recognition system. This move resulted in the deletion of face-scan data for more than one billion users. IBM also stopped offering its facial recognition technology due to these societal concerns.