News

FaceSoft Limited launches improved facial recognition technology

23 Apr 2018

Imperial Innovations, the technology commercialisation partner for Imperial College London, announces the launch of new Imperial spinout company, FaceSoft.

FaceSoft uses proprietary machine learning models and databases to improve computer-generated 3D face reconstruction and facial recognition. The technology that drives FaceSoft was developed by Dr Stefanos Zafeiriou, Associate Professor in the Department of Computing at Imperial College London & part of the iBUG group, and Mr Allan Ponniah, a plastic surgeon at the Royal Free Hospital.

FaceSoft is one of the first two companies to have launched through Imperial’s Founders Choice programme, which gives Imperial academic founders more choice regarding their service package from Imperial Innovations and resulting equity distribution. FaceSoft previously won the inaugural Programm/able software competition which provided initial proof-of-concept funding. Following spin-out formation, FaceSoft secured an undisclosed SEIS seed investment including AISeed and several angel investors.

FaceSoft uses machine learning models for 3D face reconstruction and facial recognition. The team has trained its face reconstruction algorithm parameters using a proprietary database, consisting of 2.5 million high-resolution 3D scans of real faces. The trained reconstruction model allows FaceSoft to create billions of realistic computer-generated faces, far surpassing any existing database of real faces. These avatar faces can, in turn, be used to train FaceSoft’s facial recognition algorithm, aiming to create the world's best facial recognition technology. Their recognition algorithms have already attained a >98% accuracy in the megaface competition; ranking FaceSoft top in the world when dealing with raw data alone (as opposed to clean data; see "iBUG_DeepInsight" in the leaderboards). FaceSoft’s technologies have applications in biometric security systems, medicine, and entertainment.

One key application of FaceSoft’s technology is in biometric security. Current facial recognition security applications usually rely on a photograph taken from a single reference point, such as a passport or ID photograph. In the every-day world, it is difficult to capture an image that will match with photographs of this sort, which are usually well-lit and taken face-forward. The FaceSoft algorithm can recreate an accurate model of a subject’s face in 3D, meaning that a security system can match this against images taken from various angles, in low-light, and even at low-resolution. This can dramatically increase the accuracy of facial-matching and therefore increase security.

Dr Stefanos Zafeiriou, CTO of FaceSoft, said:

“Faces are at the heart of all human interactions and connections, we are very proud and excited to have created the most accurate digital 2D/3D models of the human face and can't wait to unlock their potentials”

Alessandro Garcia, Technology Licensing Executive at Imperial Innovations, said:

“It has been very encouraging to see the winner of our inaugural Programm/able competition successfully spin-out through the Founders Choice programme and secure funding. The FaceSoft founders are a strong team composed of a compelling interdisciplinary mix. I look forward to FaceSoft’s continued progress, and how the team can vastly improve the accuracy of biometric security.”