Introducing 3D selfie
January 15, 2018
A system of Artificial Intelligence is able to reconstruct all the facial geometry in 3D from a single image, including the non-visible parts of the face.
A new era in selfies could begin thanks to a technology developed by computer scientists at the University of Nottingham and Kingston University. 3D facial reconstruction is already possible from a single image: the 3D selfie.
Through a web application, anyone can load a single color image and receive, in a few seconds, a 3D model of it. More than 400,000 users have already done the test.
The technique, while far from perfect, is a breakthrough that computer scientists have been looking for for years. What does it consist of? The new technology is based on a convolutional neural network, that is, an area of Artificial Intelligence that uses machine learning to equip computers with the ability to learn without being explicitly programmed for it.
The research team, supervised by Dr. Yorgos Tzimiropoulos, developed this convolutional neuronal network based on a huge data set of 2D images and 3D facial models. With all this information, his neural network was able to reconstruct the 3D facial geometry from a single 2D image. In addition, he was able to reconstruct the non-visible parts of the face by making a series of conjectures.
"The main novelty lies in the simplicity of our approach, unlike other techniques," acknowledges the director of the research, Dr. Tzimiropoulos.
The systems that were available until now required multiple facial images and face several challenges, such as the exact correspondence of facial poses, expressions and non-uniform illumination.
"Our neural network uses only a 2D facial image, and works with arbitrary facial poses (for example, frontal or profile images) and facial expressions (for example, smiling)," explains Aaron Jackson, another of the researchers.
The new technique is a sign of advances in deep learning, a form of mechanical learning that uses artificial neural networks (AI) to mimic the way the brain makes connections between pieces of information.
Multiple applications
In addition to the more obvious applications, such as recognition of the face and emotions, this technology could be used to customize games, enhance virtual reality, and allow consumers to try out accessories online, such as sunglasses.
Even the new 3D facial reconstruction technique could have medical applications; for example, simulating the results of plastic surgery, or it could help to understand certain medical conditions such as autism or depression.
0 comments