With the development of artificial intelligence facial recognition systems has reached very advanced levels. However, there are still some shortcomings. Especially during the COVID pandemic, the masks we wore on our faces disrupted facial recognition technologies. Although the number of people wearing masks has decreased, work on this issue continues.
Researchers from Chlef Hassiba Benbouali University in Algeria, Can recognize faces even when masks are present developed a software. The research has been published via Inderscience.
Trained with a huge database of images of people wearing masks

The researchers trained the system they developed using a large database of photos of individuals in various poses and expressions. Masks were also placed on the faces of these individuals. It was stated that the main aim of the study was to facilitate biometric identification in scenarios where mask-wearing has become the norm.
The database in question is where the masks we all once wore are placed. 18 thousand masked faces And Two thousand unmasked human faces It contained. The software uses artificial intelligence, a subset of artificial intelligence that allows analysis of visual information, to overcome the difficulty of recognizing faces hidden by masks. of convolutional neural networks (CNN) benefited. As a result of tests performed on more than one CNN model, ResNet 18 It has been observed that a model called Masked Faces can detect masked faces most accurately and effectively.
As we know, facial recognition systems can be used in many different areas, from unlocking phones to tracking down criminals. Researchers also say the technology they developed can be integrated into these areas and could be very useful in mask use.
Despite the promising results, the limitations of the system and how it will perform in practice are unclear. Furthermore, although ResNet 18 is successful, it is unknown whether this model will overcome difficulties in different situations, such as different lighting conditions or changing mask types. That bad to further research there is a need.
Follow Webtekno on Threads and don’t miss the news