What is facial recognition? How facial recognition works?
The Definition of Facial Recognition
Did you ever wonder how you were able to use Apple Face ID as a phone lock? Or how in action movies, law enforcement was able to detect someone's identity by just uploading a picture? Here you will learn what facial recognition is and how facial recognition works.
Using various technologies, facial recognition identifies human faces. When a picture of a human's face is matched to other known faces in a database, facial recognition can be performed. Facial recognition may help identify a person; however, it can create several issues regarding privacy. Briefly, facial recognition can be defined as the pre-practice of identifying human faces from a real-life video or a camera picture.
Having scalable application areas varying from marketing to surveillance, the facial recognition market is expected to grow to 8.5 billion by 2025, according to research done by Markets and Markets. However, this promising technology has over 50 years of history behind it.
To begin with, the pioneers of facial recognition technologies are Woody Bledsoe and Helen ChanWolf. They were the first to use a computer to recognize a human face, and they started doing so by manually pinpointing the facial feature: the inner and outside points of eyes, the centers of pupils, the hairline's widow's peak, and so much more. Afterward, these facial features were used to calculate distances to decide the width of the eyes and mouth. These findings were stored in a database, and for each picture, the computer would automatically match the closest pairs.
Between 1970 and 1997, it was demonstrated by Takeo Kanade that the chin and the distance between facial features could be identified, but with later test results, trust in the accuracy of the facial recognition process declined. Nonetheless, until the 1990s, the governments supported and helped spawn new tech companies focused on facial recognition.
The more reliable facial recognition method was proposed by Matthew Turk and Alex Pentland in the early 1990s. This method was called Principle Component Analysis (PCA) and drastically transformed the future of facial recognition. Before Principle Component Analysis was developed, it was hard to perform facial recognition in a picture that contains objects other than the face. The method is known as Eigen's face as well, combined factor analysis and Karhunen - Loeve theorem in order to develop a linear model. The face of a human was computed as a weighted combination of Eigen's faces.
Following advancements in the field by contributions of many, real-time facial recognition occurred in2001 via the framework called Viola-Jones object detection. This framework enabled the implementation ratio of facial recognition in personal portable devices and embedded systems. Afterward in, commercial usage and new features in technology for facial recognition systems started.
Moving to the 2010s, facial recognition systems started to provide more thanks to advancements in computers. Governments and agencies started to use facial recognition systems to identify persons of interest. Furthermore, around 2015s, companies started to use facial recognition in their smartphones as a method for face authentication. Android's Trusted Face, Apple iPhone X's Face ID, and windows Hello can be given as an excellent use facial recognition systems in mobile phones. Last but not least, it is said that the IT Section of Ukraine's army has been using facial recognition systems to identify deceased people during the war.
How Facial Recognition Works?
There are several techniques combined with technologies that enable humans to perform facial recognition; however, facial recognition is still a big challenge for computers. Unlike humans, computers are not able to process a picture and identify individuals within the blink of an eye. Yet, the process is still familiar to how humans identify a person. Humans see eyes, mouth, nose, and hair, whereas computers see only pixels.
To begin with, a picture or an image consisting of at least one human face should be collected. These pictures can be taken in the form of a photo or a video. There can be multiple faces or just one human face on the taken picture. Almost in profile or looking straight ahead, images can be applied for a facial recognition system to work. After the picture is taken, the face is segmented from the picture's background.
Secondly, alignment of the segmented face must occur. This alignment will enable to the description of image size and properties of the picture, such as illumination and grayscale. The reason behind the alignment is to aid the third step, which is facial feature extraction.
As the next step, the facial features should be extracted from the segmented image. The pinpointed and measured properties of the facial features are eyes, nose, and mouth, respectively. To elaborate, the distance between the eyes and the distance from the forehead to the chin are pinpointed and measured. Afterward, these facial features are stored and held for the final step.
The algorithm running the facial recognition system should compare extracted facial features to a database of known faces and come up with an accurate result. In this final step, the extracted facial feature vector is matched with a known face.
Other Facial Recognition Technologies
Even though the majority of facial recognition systems work similarly as described, there are still other techniques that can be used for facial recognition. The first one is called 3-dimensional recognition, a technique that is not affected by the transformation of light. This technique collects information about a face's shape using 3D sensors. The information about the shape of a face gathered is similar to the extraction of facial features. One other advantage of 3 Dimensional facial recognition is that this technique is sensitive to expressions.
Secondly, for facial recognition at a distance, there is a specific technique called Human identification at a distance. This technique enhances the resolution of low-resolution images via face hallucination. In Human Identification at a distance technique, pictures are processed via a machine learning model before being sent to facial recognition systems. During the machine learning process, pixel substitution or nearest neighbor distribution might occur. The advantage of this technique is that it overcomes the constraints of pictures taken at a distance.
Last but not least, there is another technique called thermal camera detection, which captures the shape of the head without any accessories such as hats or makeup. In low-light and nighttime conditions, thermal cameras are known to be working effectively. Yet, the database is limited due to the restricted usage of thermal cameras.
It is a known fact that facial recognition systems are being used rapidly in traveling, banking, insurance, and even in soft drinks! High tech companies, such as Apple and Microsoft, are working toward creating better facial authentication for their consumers, and other companies are following in their footsteps in adopting these technologies. For instance, in the automation industry, it is known that Intel and Ford are testing driver authentication for identifying the primary driver of the car. Coca-Cola has been using the rewarding program for recycling in China and has been using facial recognition technologies for some special rewards.
There is a lot of controversy concerning privacy, mistaken identity, and the accuracy of facial recognition systems. To elaborate, the detected faces stored in a database are at the mercy of law enforcement agencies or other retailers. This is indeed an invasion of privacy. To add, since the accuracy of these facial recognition technologies might differ, there are several cases in which the identity of the detected person is mistaken. Criminals, in fact, create new ways that can trick facial recognition systems. However, the pros of facial recognition technologies can not be undertaken. Being able to find a missing person, identifying the right criminal, making authentication to phone easier, and making banking available within the blink of an eye are the indispensable uses for facial recognition systems. Furthermore, with developments in facial recognition systems, emotion recognition can be applicable and can help identify whether a person of interest is lying or not. As much as facial recognition systems work towards identifying faces, anti-facial recognition systems are working for the exact opposite purpose: to anonymize the faces!
Cameralyze Facial Recognition Technologies
Nonetheless, it is unavoidable that facial recognition technologies will be rapidly applied in a variety of ways, and if you are interested in using facial recognition in your business, make sure to check out Cameralyze's face blurring and face recognition solutions!