Being living in the 22nd century the AI game in our day to day life has raised to a benchmark and Facial recognition is one of the most prominent ways to have your digital content secured. After all Privacy is a must! But with coronavirus in the scene life has went upside down and so is the need for the technological advancements has raised.
Having face mask as a prominent day to day accessory protecting us from Covid-19 virus it has also came up as an obstruction for the facial recognition system. It’s because these system works on mapping distance between eyes, nose, chin, and mouth to identify a person and having half of our face covered with masks it is impossible for the system to create an accurate facial signature thus affecting the purpose of facial detection system.
This has come up as a major challenge for developers and researchers. The National Institute of Standards and Technology (NIST) conducted a study on existing algorithms to figure out the effectiveness and solution to cope with this new situation.
Here’s the findings of study.
89 algorithms have been tested on images of million people collected from the surveillance system at US border and immigration applications. Digital masks of various shapes and colors were put on the images to mimic the scenario.
Around 5-50 percent error rates were found in the results based on different color and shape of masks. Results were positive for those experimented with round masks whereas those with dark masks gave less accurate results.
These findings show there is a need to update the current face recognition algorithms. Panasonic, Facewatch, SenseTime, and few more companies have already started looking for solutions to this problem since February.
And the researchers and developers have found several ways as a solution to this.
Face On The Mask
One such company of US, Resting Risk Face came up with a unique idea to ease the process for facial recognition systems.
The mask order is still in effect here in Colorado, so what masks have you been rockin’? Would you get a mask of your face like this one Danielle Baskin designed, or do you keep it simple?#facemask #maskup pic.twitter.com/LbJShuPChe
— 20/20 Tax Resolution (@2020TaxResInc) September 3, 2020
Danielle Baskin, artist and owner of company used the face printing technology to solve the issues. She made a 3D image on the mask that help depth sensors of phone to recognize the face and unlock the phone.
Future Generation Computer Systems published a study where machine learning is used to find out the recognition rates of computer when face is partially or quarterly visible.
A deep learning technique known as convolutional neural network was used along with VGG model for facial recognition. The model was tested for different visibility of facial images.
For full visibility, results were 100% accurate whereas for partial facial images only 90 % accuracy was achieved with non-visibility of eyes and nose. On the other hand, recognition rates were low for other facial areas like forehead, cheek, and mouth.
This experiment has given a possibility that if existing models work with partial images then facial recognition would be better even when with masks on.
Biometric Authentication Terminal
NEC has developed Bio-IDiom system, a touchless model that uses face and iris detection, to improve the capability of facial recognition by combining two different technologies.
It adjusts the light and focus on the face to authenticate the face profile of an individual. This technology is said to be highly effective by the NEC as it gave more accurate results than other algorithms during NIST testing.
Indigenous facial recognition technology, FacEDGE is said to be 95 percent accurate, fastest, and reliable. It is powered by EDGENeural.ai which is a Pune based startup. Company’s co-founder, Sarvesh Devi said the technology is effective against the face masks.
So, having the technology as a savior no virus can refrain humans from being on point in their tech game.