Face recognition new technology accuracy exceeding 99%: more accurate than the naked eye.

In the early hours of June 23, Tang, Wang Xiaogang, a professor at the Chinese University of Hong Kong, and his team announced last week that they had developed a deepid face recognition technique that was more accurate than 99%. The Computer Vision Research Group (MMLAB.IE.CUHK.EDU.HK), led by Tang, developed a depth learning model called deepid (deep), with a 99.15% recognition rate on the LFW (labeled JavaServer in the Wild) database. LFW is the most widely used test datum in the field of face recognition. The experimental results show that the recognition rate of human eye on LFW is 97.52% if only the center region of the face is given. Prior to this, Tang's research team developed a face recognition technology based on Gauss process Gaussianface (Gauss face), achieved 98.52% recognition rate. This is also the recognition rate of computer automatic recognition algorithm for the first time more than the naked eye. Deepid the Gaussianface's face recognition world record and pushed forward a step further, for the first time more than 99% LFW recognition rate. Face recognition is an important challenge in the field of computer vision and artificial intelligence, which is widely used in the fields of public security, law enforcement, mobile internet and entertainment. It has also become an important test benchmark for testing whether AI can reach or even surpass people in solving certain intelligent problems. Tang's research group has more than 10 years of research experience in the field of facial recognition. They began the study of deep learning methods in 2011, reaching a 92.52% recognition rate in 2013. In the past year, they have raised this figure to 99.15%, and earlier this year, Tang and Wang Xiaogang's team had released a face recognition algorithm based on depth learning, which achieved the highest 97.45% recognition rate on the LFW. At the same time, Facebook (64.5,0.16,0.25%) has released another set of face recognition algorithms based on depth learning DeepFace, with a 97.35% recognition rate on LFW. DeepFace needs more than 7 million face data for training. The deepid uses only 200,000 face data and several nvidia (18.93,-0.21,-1.09%) K40 GPU. At present, the Tang Lab's three face recognition algorithms occupy the top three LFW recognition rate, while Facebook's DeepFace ranked fourth. Tang that there is a lot of work to be done in face recognition field, many algorithms need to be improved and improved in practical application. His lab has created a complete set of human face image processing system SDK based on the latest technology breakthroughs, including face detection, face key point alignment, face recognition, facial recognition, gender recognition, age estimation and other basic technology packages. Tang plans to provide face recognition technology free to Android, IOS and Windows Phone developer; With the help of this set of FREEFACE-SDK, developers can develop various applications based on face recognition on mobile phones. In addition, Tang also hopes to use user feedback to further improve the accuracy of the algorithm. In addition to face recognition, Tang and Wang Xiaogang's research group's other core research direction is deep learning. They have designed a number of in-depth learning models that can be used to study many important problems in the field of computer vision, including face alignment, pedestrian detection, attitude estimation, human body image segmentation, vehicle recognition, large-scale crowd monitoring, universal object recognition and detection, and Internet image retrieval. Deep learning is regarded as the biggest breakthrough in the field of artificial intelligence in the past ten years, and it has many applications in the fields of computer vision, speech recognition and natural speech processing. MIT Technology Review included the 10 most groundbreaking technologies of the 2013. Deep learning attempts to imitate how the human brain uses neural networks to perceive the world. Its results largely benefit from large data and parallel computing based on GPU in recent years. Baidu (174.5,0.11,0.06%) established a deep learning Institute in 2013, and in December, Facebook created a deep learning AI lab in New York. 2014, Google (556.36,1.46,0.26%) 400 million U.S. dollars acquisition of in-depth learning start-up company DeepMind Technologies. (Yan Fei)