Bulletin No. 2, 2014

Scientists, Muses and their Labs  21  Taking Face Recognition to New Levels Prof. Xiaoou Tang (top), professor in the Department of Information Engineering, and Prof. Xiaogang Wang (middle), assistant professor in the Department of Electronic Engineering, have built a novel facial recognition s y s t em t h a t h a s t h e highest accuracy in the wo r l d . Wh i l e h uma n beings generally recognize faces with an accuracy r a t i ng o f 97. 53% o n L a b e l e d Fa c e s i n t h e Wild, a database of face photographs designed for studying the problem of unconstrained face recognition, the CUHK-developed system is observed to recognize faces with an accuracy of 99.15%, regardless of changes in lighting, make-up and camera angles. The project was carried out by the CUDA Research Centre, Hong Kong’s first NVIDIA CUDA Research Centre, which prepares engineers and computer scientists to conduct potentially groundbreaking research using GPU (graphics processing units) accelerators. NVIDIA is an American global technology company that manufactures GPUs. The system developed by the professors and their team has benefited from deep learning, one of the biggest breakthroughs in artificial intelligence in recent years. Simply put, deep learning is a sophisticated ‘machine learning’ algorithm with abilities to recognize syllables and images. It provides the CUHK system with powerful tools to improve the accuracy of face recognition. The system has many potential applications. For example, it can help law enforcement and security agencies seek out individuals among a crowd of thousands. Traditional video surveillance can only focus on a small number of objects in a very simple environment, but the new system allows its users to target thousands of objects in very complex environments. ‘ The key challenge of face recognition is to develop feature representations that reduce intra- personal variations and enlarge inter- personal differences. Deep learning enables our system to handle the two types of variations effectively,’ said Prof. Xiaogang Wang.