Computers are nowadays so involved in our lives that we cannot imagine a life without them. Computers are actually excelling at the tasks of the humans. Nowadays computers are actually doing better at tasks that are uniquely human. Artificial Intelligence is just thriving nowadays. One of those many tasks is facial recognition, a task that was dominated by the humans for a long time, but now computers have outperformed the humans in this task as well. Computer scientists have recently developed an algorithm that performs the task way better than the humans.
Facial recognition systems were always good at comparing two images in proper light conditions to check if it’s the same person or not, but only when you shake up those conditions a little, such as lighting, poses and expressions, humans were seen to perform a better job as compared to the computers, actually at an accuracy of 97%. Until now, no facial recognition algorithm was able to cross that accuracy percentage, but recently scientists have developed a new algorithm called GaussianFace, that accomplished an accuracy of 98.52%.
This algorithm breaks up the face it sees into small portions, and verifies the position of the eyes, nose and corners of the mouth in consistent locations. Then, it further divides the image into small portions, overlapping each other and charts the unique characteristics of each portion, allowing it to compare two images.
Once the algorithm was tested and trained on the test subjects, it was time for the big league. The LFW (Labeled Faces in the Wild), that is a dataset of 13,000 facial images, that may or may not be of the same person and have the conditions shook up, that is the lighting, makeup, angles and expressions are all switched up. This is the ultimate test for any facial recognition algorithm.
The GaussianFace algorithm outperformed the humans in this test. Beating humans’ 97.35 percent average performance on LFW. The future aspects of this algorithm are amazing. Starting from your cell phone’s protection and widening up to even National Security. These findings were published by the scientists in the online journal arXiv.
Want to check how you fair against the algorithm? Click the link below: