Dahua Technology reports a new record for Labeled Faces in the Wild (LFW, Facial Recognition, beating the record set previously by Baidu, Tencent, Google and others.
Recently, the facial recognition technology team from Dahua Technology presented their results for Labeled Faces in the Wild (LFW), which is a worldwide facial recognition database test set. With a series of technical improvements, Dahua reports that its FaceImage facial recognition system was not only ahead of Google, Facebook, Baidu, and Tencent, it also set a new LFW record.
LFW was established in 2007 by the University of Massachusetts and is used to evaluate the performance of facial recognition algorithms under unconstrained conditions. Up until now, a few dozen teams from around the globe have provided more than 80 test results. These have come from companies such as Google, Facebook, Microsoft Research Asia, Baidu, Tencent, SenseTime, Face++, Chinese University of Hong Kong and others from industry and academia.
In recent years, most of the improvement in facial recognition technology have come from the development of deep learning. That’s a machine learning approach that simulates the human neural system. The functions of Convolutional Neural Networks (a commonly used deep learning model) are closely related to network depth. However, since deep networks are difficult to optimise, the network depth for facial recognition methods of the past generally range from a few to a few dozen layers. Dahua reports that its facial recognition technology team has designed a network with a depth of over 100 layers (the deepest network layers among facial recognition system so far announced). The firm says this enables a new type of metric learning method which allows the similarity score to be higher for images of the same person. At the same time it lowers the similarity score between images of two people. Combined with a sampling technique, the rate of convergence can also be increased, the company adds. By training multiple models and using a non-linear multi-model integration technique, Dahua reports an accuracy of 99.78 per cent for the LFW dataset.
Dahua says that its facial recognition technology team, part of the advanced technology research institute in its R&D centre, supports Dahua’s facial recognition related products (face detection, face feature point location, face recognition, face attribute analysis, smile detection and so on) for commercial use.
A spokesman for the facial recognition technology team, Prof Wang Haiyang said that achieving high accuracy in open data sets had motivated his team to work on the more challenging implementation in real-life situations. Dahua has accumulated a huge volume of video data. By using this data to adjust the learning model, the performance of the algorithm in real-life situations has been greatly improved. Furthermore, the security application requires the recognition algorithms to respond quickly. The technical team used network pruning and a multi-model feature sharing technique to greatly reduce redundant operations so that the amount of calculations required for networks with over 100 layers is similar to that for networks with only a few dozen.
Dahua adds that its facial recognition has already been applied to public security, finance, and other areas. At the latest G20 summit, in the Chinese city of Hangzhou, Dahua’s facial recognition technology was applied. Dahua cameras were deployed in core G20 activity areas and in many traffic hubs. Real-time images captured were automatically matched with security black-lists in the back-end. When a high-risk person showed up in the video, the system can issue a warning message to a command centre to dispatch response. The firm reports that this system played an important role during the summit and assisted police in a number of arrests while the system was still in its commissioning phase.
