6 August 2014

Novel Face Recognition System Developed by CUHK Faculty of Engineering Achieves 99.15% Accuracy



A CUHK research team led by Prof. Xiaoou Tang, Professor, Department of Information Engineering, and Prof. Xiaogang Wang, Assistant Professor, Department of Electronic Engineering, has built a novel facial recognition system that with the highest accuracy in the world.  While humans recognize faces at an accuracy rating of 97.53% on Labeled Faces in the Wild, the recognition system developed by CUHK was tested using thousands of picture sets, and it can recognize faces at an accuracy of 99.15%, regardless of changes in lighting, make-up and camera angles.  This is the first time for computing algorithms to reach human face verification performance on this dataset. 

'The key challenge of face recognition is to develop effective feature representations for reducing intra-personal variations while enlarging inter-personal differences,' said Prof. Xiaogang Wang.  'With deep learning, the system is provided much more powerful tools to handle the two types of variations and significantly improves the accuracy of face recognition.  This technology has numerous important applications in security, law enforcement, Internet and entertainment.'  The system could help law enforcement and security agencies to 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.  With the new system the users can target thousands of objects in very complex environments. 

Deep learning is the biggest breakthrough in artificial intelligence (AI) in recent years.  It simulates human brain's behaviors by training large scale neural networks from big data based on intensive graphics processing unit (GPU) computing. 

Face recognition is one of the most important grand challenges in computer vision and AI.  Scientifically, this is also an important benchmark on whether AI can reach the level of human intelligence or even surpass it. The breakthrough achieved by CUHK is a strong evidence that deep learning makes AI possible. It opens the door to many important applications, such as finding terrorists from surveillance videos, recognizing imposters at ATM machines, and automatically tagging face images uploaded to social networking sites. 

The CUDA Research Center at CUHK 

As a pioneer in the field of deep learning, CUHK has been selected as Hong Kong's first NVIDIA[1] CUDA[2] Research Center which aims to prepare researchers, engineers and computer scientists for ground-breaking work using GPU accelerators. Prof. Xiagang Wang is the director of this new CUDA Research Center. CUHK will utilize the facility and technical support provided by NVIDIA to enhance its computing arsenal in the areas of deep learning.  With the support of GPU parallel computation systems, researchers at CUHK will continue to develop deep learning technologies and apply them to various computer vision related applications including video surveillance, web scale image and video search, as well as human and computer interaction.  The Department of Electronic Engineering at CUHK will also offer a first-ever graduate course on deep learning in the 2014-15 academic year to nurture local talents in GPU related applications. 

The new CUDA Research Center at CUHK will provide various facility and technical support for the University to conduct GPU related research activities.  CUHK will have priority for pre-release access to hardware and software provided by NVIDIA.  Recently, NVIDIA has provided the fastest Tesla cards to support CUHK's research of deep learning on face recognition. It normally takes one month for a central processing unit (CPU) to train a deep neural network for face recognition, while a Tesla K40 GPU can complete the training process within 10 hours. NVIDIA will provide GPU training and education sessions for engineers and researchers at CUHK.  The engineers from NVIDIA will help the research groups at CUHK set up the optimal configurations of GPU computing systems for crowd video surveillance and training of deep neural networks. They will also help re-implement the computing algorithms developed by CUHK to improve their efficiency on GPU and make them assessable by other GPU users, which is very important for generating impact of research at CUHK. 

[1] About NVIDIA®

NVIDIA is the world leader in visual computing technologies and the inventor of the GPU, a high-performance processor that generates breathtaking, interactive graphics on workstations, personal computers, game consoles and mobile devices. 

[2] About CUDA® (Compute Unified Device Architecture)

CUDA is NVIDIA's parallel computing platform and programming model that enables dramatic increases in computing performance by harnessing the power of GPU.  Institutions identified as CUDA Research Centers are doing world-changing research by leveraging CUDA and NVIDIA GPUs.

The facial recognition system developed by CUHK conducts face parsing via deep learning.
The facial recognition system developed by CUHK conducts face parsing via deep learning.

Prof. Xiaogang Wang
Prof. Xiaogang Wang

Prof. Xiaoou Tang
Prof. Xiaoou Tang