VietNamNet Bridge - Even with face masks or beards, artificial intelligence (AI) can recognize faces at a high accuracy rate of 86 percent, according to a study "Identification of human faces through inadequate information" by Vo Hoang Trong, a Mathematics and Information Technology major from HCMC National University.


{keywords}


When talking to friends, Trong often heard statements identifying features such as “the man has big eyes” or “I guess he is so and so because there is a scar on his forehead”. This prompted Trong to think of a biometric face recognition solution to determine the identity of individuals.

FaceID technology on Apple's phones builds a three-dimensional model of human faces based on sensors, which makes it easier to identify different sides and hidden faces.

However, the technology cannot be used for public cameras because it cannot always scan the three-dimensional face of a person in the crowd.

Trong decided to use AI to recognize human faces in different conditions, even when images have low quality.

Trong thinks that if the information about a face is insufficient (there is only information about eyes, nose and forehead), it would be better to identify areas on the face separately to discover typical characters, such as ‘big eyes’, or ‘high nose’. 

Trong thinks that if the information about a face is insufficient (there is only information about eyes, nose and forehead), it would be better to identify areas on the face separately to discover typical characters, such as ‘big eyes’, or ‘high nose’. 

After that, it is necessary to combine different identification results to discover the identity of the person in the image.

Tests show that Trong’s identification method has a high accuracy rate of 86 percent if the information about faces is sufficient. 

The accuracy rates would be 80.87 percent if there is only information about forehead and eyes (in case person has a face mask), 73 percent if there is only information about nose and cheeks, and 64.87 percent if there is information about mouth and chin. For the images of the faces with only the left or right cheek, the accuracy of the algorithm could be up to 57.5 percent.

Solution for smart city

Trong asserted that the solution can applied to smart cities. Many smart urban areas have used the solution to recognize human faces in the crowd through security cameras, or identify persons going in/out of companies.

He suggested using individuals’ images on identity cards, student cards and passports to train computers, so that they can recognize faces later.

“Those trying to commit illicit behaviors tend to disguise themselves by wearing artificial beards or hair. However, with my solution, AI can find out who they are,” he said.

Pham The Bao from the HCMC National University praised the invention, saying that it should be announced in ISI publications.


RELATED NEWS

BKAV runs second successful mask test fooling Apple’s Face ID

VN student wins US invention prize


Mai Chi