No one remain hidden

You must have known that the first major legal challenge in the UK over police use of automated facialrecognition surveillance has begun in Cardiff just yesterday. In other side, police in London conducted a public street trial with facial recognition cameras. A man who covered his face as he walked by the cameras was stopped by officers, forced to submit to being photographed, and then arrested on a charge of public disorder after complaining loudly.

In US, opposition to Amazon's sale of its facial recognition technology to US police forces is set to come to a head at its annual general meeting on Wednesday. San Francisco, long at the heart of the technology revolution, took a stand against potential abuse on Tuesday by banning the use offacial recognition software by the police and other agencies.

We can really see that the mass facial recognition technologies  are facing serious controversies as opposed to China, where social credit system relies on facial recognition on surveillance cameras as citizen control tool. So people are going nuts about facial recognition,but what if they see a newer technology like "Facial Unmasking"? Yeah, that sounds correct! In the past, it has been thought impossible to uncover a covered face from single image with pretty enough details to perform facial recognition, but in the age of Artificial Intelligence, the fiction has become reality.

SIGMIND, a Bangladeshi startup has been working on cutting edge computer vision technologies powered by A.I. to automate the surveillance industry since 2016. Our WATCHCAM software suite is being used by the government of Bangladesh to secure some Key Point Installations (KPI) premises. The software can recognize subjects by their face and also identify their intrinsic features like color of clothing. It also detect and recognize vehicles by their license plates. An example of the use case is the government owned IT incubator named “Software Technology Park” located in Dhaka where each and every person is tracked by surveillance cameras by facial recognition as well monitor and vehicles on the parking by license plates. The system also monitor attendance and provide access control functionalities.

But as we always looking for something groundbreaking, here at SIGMIND, we have developed facial unmasking technology that can generate realistic enough face details to recognizes faces even if it's masked or covered partially! The system use the latest advancements in computer vision technologies to learn facial features from millions of high definition face images and generate realistic facial details from single image on the fly. The module has also been integrated in the core WATCHCAM software so that no-one could hide in the eyes of cameras.

The system detects faces from surveillance cameras and classify is it is masked or not. If masked, the cropped face image is fed into unmasking pipeline where, half of the face is blanked out and then a pretrained deep learning model tries to generate the whole face from the half face. After generation, the full face is then fed into recognition pipeline where is is matched against thousands of database images by 128 point facial landmarks. The unmasking system has been tested on the Labeled Face in the Wild (LFW) database popular for bench-marking facial recognition, and amazingly matched 82% faces correctly from only 256x256 size face images.

We think such technology could revolutionize the the surveillance industry but at the same time may raise controversies that happens to all new technology inceptions. In this severe time, if such tool like facial unmasking is integrated with the existing facial recognition systems, criminals who prefer to commit crime by covering their face will be discouraged. For example, bank robbers wear masks during robbery most of the time to defy the security cameras but no more if this technology is deployed in those systems.

This technology could also be applied to forensic analysis or criminal investigation from recorded footage not only, this could be deployed in in real-time to capture suspects in more rigorous way like from helmet/mask wearing subjects. So this is evident that the battle between technology and government is is eminent that's why the regulations are becoming more stringent.

We have thought this might be an interesting topic to cover, as the West is getting confused with the promise of facial recognition, in Asia, a computer vision startup has been developed something extraordinary in that industry. Its also a sign of the democratization of AI, where not only big corporations, but small startups also using it to invent and revolutionize the industries. The promises of this technology could also be abused in ways like Deepfake, thus we have to ensure its in right hand.

SIGMIND has also been on NVIDIA Inception Program since 2017 which recognizes top AI startups around the world. The company has been funded by the ICT Division from the government of Bangladesh and has been received National ICT Award for its extraordinary works. Recently, the company has been participated on the Startup Turkey conference and listed on the Top 10 promising startups.