Facial Recognition - How Does it Work?

Facial or Face recognition has lot of advantages and applications in modern technology world. One of the common and rather latest examples is use of facial recognition to access Smartphone. Apple was one of the first to introduce facial recognition system called Face ID in iPhoneX. It is now estimated that over billion smart phones will have facial recognition system by 2020. We have also seen facial recognition technology being used in thriller or detective movies where CCTV footage is used to identify a suspect and to establish the identify his photograph is run through a database of criminals to find a match using facial recognition technology. There are numerous other applications of this technology as well. This article discusses very briefly how facial recognition system works. There are two types of facial recognition environments - Constrained and Unconstrained

Constraint environment is the environment where images are collected with careful cooperation of the persons or subjects, therefore face recognition is more accurate. Figure 1 shows examples of constrained environment. Constrained environment is also known as controlled environment in which we can control factors that affect face recognition like illumination, pose, expression etc. For example Person stands in front of a camera where illumination conditions are good is an example of constrained environment.

Figure 1: Constrained Environment

In unconstrained environment images are collected without person or subject cooperation and subject is generally not aware about the fact that his/her image is being collected for facial recognition. There are a lot of factors like illumination, pose, expression, age etc. that affects face recognition. Hence it is obvious that in unconstrained environment accuracy of face recognition is much lower as compared to the constrained environment. Factors like compression of images, obstruction, shadows, blurring etc result in lower accuracy. Persons or Subjects with different pose or expression style, wearing sunglasses, scarf, images of crowd, candid images etc. come under unconstrained environment as shown in figure 2.

Figure 2: Unconstrained Environment

As shown in Figure 3 facial recognition system include face detection, alignment, feature extraction & feature matching. An image of a person or subject is sent to a face detector that will identify the face and check the alignment. Face alignment is done to align the face as per the requirement of the application. Using different algorithms feature extraction is done. Features may include eyes, nose, ears etc. The extracted features are matched against the test data or test images in the database.

Figure 3: Facial Recognition System

Face recognition systems through face detectors are capable of achieving satisfactory results in constrained environments however when it is applied in unconstrained environment the detection accuracy falls to 50-70%. In addition to this approximately 3% faces are detected false positive. A lot of work is being done in this field especially to improve accuracy in the unconstrained environment.

By: Mr. Chetan Sharma - Assistant Dean (CSE), Chitkara University, H.P.

References

1. www.nist.gov
2. Abate, A. F., Nappi, M., Riccio, D. and Sabatino, G., 2D and 3D face recognition: A survey, Pattern Recognition Letters, 28(14):1885-1906, 2007.
3. Zhao, W., Chellappa, R., Phillips, P. J. and Rosenfield, A., Face recognition: A literature survey, ACM Computing Surveys, 35(4):399-458, 2003.
4. Amrit Kumar Agarwal, Yogendra Narain Singh, Evaluation of Face Recognition Methods in Unconstrained Environment, Procedia Computer Science 48(2015) 644-651