Introduction
Technology has been through tremendous development over the last decades. Many aspects of life have changed due to new technological advancements. A new system that has been used in many fields is facial recognition. It is a tool that can match a face to an existing database using images and videos. Typically, facial recognition is used in social media, ID verification, security services, and other fields. While facial recognition seems to be a helpful practice in various situations, specific experiments suggest that it has downsides and flows. Apart from the ethical concerns of facial recognition being a practice that invades privacy, there is data that proves the incapacity and impairment of this method. Depending on inevitable distortion, the results are much less impressive than initially intended.
Experiments
Since facial recognition is a topic that has been under discussion for the last couple of years, there have been multiple experiments that studied the efficiency of this system. The two experiments that show how limited the current potential is are “Deep face recognition using imperfect facial data” (2019) and “A real-time face recognition system based on the improved LBPH algorithm” (2017). The researchers analyzed the criteria that created an advantageous recognition system by studying different types of images and how the system interacts with them. Since facial recognition works by comparing a given image of an individual to a database where this person is present, an efficient outcome would be matching the two pictures 100% of the time. However, the experiments showed that certain distortions make it hard or impossible for the system to work as intended.
Real-Time Face Recognition
Real-time face recognition is often used in criminal investigation, authentication, and surveillance. This is a widely used method that is used on a day-to-day basis. It has become a standard technology for smartphone owners who can add a facial recognition security system to their devices. While this can be a useful tool, it has been noted that depending on certain factors, facial recognition may not be nearly as efficient as expected. Researchers were able to conclude that factors such as illumination, rotation, and proportions play a crucial role in recognition efficiency (Zhao & Wei, 2017). Depending on how bright the light is, how close the person is to the camera, and how straight the face position is, the system may or may not show impressive results. It has been concluded that the percentage of correct matches can be very low depending on these aspects of real-time facial analysis.
Illumination
Illumination plays a significant role in the efficiency of the facial recognition method. According to the findings of researchers, dim lighting severely reduces the potential of such systems (Zhao & Wei, 2017). The experiment proves that bad lighting causes the system to be correct about 30-40% of the time, which is a low number. This proves that lightning can be a major factor that distorts the results.
Rotation of the Face
Rotation of the face is also an aspect that may be difficult for facial recognition systems to cope with. According to the experiment conducted by Zhao and Wei, the position of the head can majorly distort the results (2017). The investigation shows that a face rotated 30 degrees to the right or left reduces the recognition rate to 40-50%. This means that a slight change in the position of the head makes it less likely for the system to make a correct assessment.
Face Proportion
Depending on how close or far the face is from the camera, the proportions are distorted. According to researchers, the closer the face is to the camera, the more serious the distortion is. As suggested by the experiment, a person that is closest to the camera is much harder to recognize by the system, with a recognition rate of only 20-35% (Zhao & Wei, 2017). This means that an individual that stands too close to the lens is less likely to give the facial recognition tool valid data for comparison.
Imperfect Facial Data
While facial recognition proves to be effective, the data must be impeccable for the system to work as intended. In other cases, there can be significant problems. According to the experiment conducted by researchers in 2019, imperfect facial data correlates with poor performance. It has been noted that partial faces, rotated images, and zoomed-out photos are challenging for the system to analyze and compare to the information in the database objectively.
Rotated Faces
Distorting a face by rotating the image proves to be an efficient way to test the limits of facial recognition systems regarding data comparison. Researchers came to the conclusion that rotating the face image between 110 and 120 degrees results in poor recognition. In some cases, the rate was 0% (Elmahmudi & Ugail, 2019). This proves that facial recognition is almost impossible in the circumstances of the image being distorted with rotation.
Partial Faces
Using images that portray only certain parts of the face has shown that recognition is not as effective. While the eyes seem to be the only facial part that is recognized by the system in around 65% of the cases, the nose, forehead, and mouth are the worst parts to be analyzed. The results tend to be close to 0%. Showing half of the face also distorts the percentage of exact guesses (Elmahmudi & Ugail, 2019). This means that people that wear face masks, head scarfs, hoods, and glasses are less likely to be recognized by the system.
Zoomed Out Faces
Pictures that are zoomed out seem to be almost impossible to use for facial recognition. Depending on how many percent the photo was zoomed out, the results of the experiment differ. However, 60% of the zooming level is enough for the recognition rate to drop from 100 to 60%. The efficiency is even worse for 70-90% of zooming. The rate of reconditioning drops to as low as 0% (Elmahmudi & Ugail, 2019). Zooming an image out proves to be a distortion that the system is not yet advanced enough to work with.
Conclusion
Facial recognition is widely used in different fields. Although the system is helpful in some instances, there are multiple factors that may induce errors. The system loses its efficiency based on the lighting, position of the face, proportions, rotation, partial images, and zoom. These are the distortions that can cause the efficiency rate to drop dramatically. While a human can match a rotated image of a side-ways picture of the individual’s profile, technology is not as advanced at the moment. There is definitely room for improvement when it comes to facial recognition, but based on the experiments mentioned above, it is not yet advanced enough to be called a fully efficient system.
References
Elmahmudi, A., & Ugail, H. (2019). Deep face recognition using imperfect facial data. Future Generation Computer Systems, 99, 213–225.
Zhao, X. M., & Wei, C. B. (2017). A real-time face recognition system based on the improved LBPH algorithm. 2017 IEEE 2nd International Conference on Signal and Image Processing (ICSIP).