Deep learning subverts traditional biometrics

HC Security Reuters currently used in the field of security is more common biometric identification technology on the part of the identity features include computer vision-based technology, in addition to the human face, gait, involving authentication as well as fingerprint, iris, palm prints, Veins, etc., all rely on computer vision technology for image pattern extraction, content analysis and recognition. There are also biometric recognition technologies such as speech recognition, brain wave recognition, and saliva extraction DNA.

In recent years, artificial intelligence is very lively. If 2016 is the "first year of artificial intelligence", then from 2017 onwards, it will enter the era of mature artificial intelligence. In 2018, it is possible to usher in the application of artificial intelligence to industry segmentation. Year.

Deep learning subverts traditional biometrics

The concept of "deep learning" as an artificial intelligence engine was proposed in 2006, stemming from research derived from artificial neural networks, by combining low-level features to form more abstract high-level representation attribute categories or features to discover distributed feature representations of data.

From the perspective of current industry development, deep learning, as an algorithm that mimics the external world of human brain neurons, has made great progress in image, sound and semantic recognition, especially in the field of image and sound, which greatly improves the recognition rate compared with traditional algorithms. .

Image recognition: Image is the earliest field of deep learning. Some well-known artificial intelligence experts have obtained some small-scale (hand writing) image recognition results through the study of convolutional neural networks until 2014, The Chinese University of Hong Kong The computer vision research group led by Professor Tang Xiaoou developed a deep learning model called DeepID. In LFW (Labeled Faces in the Wild), face recognition uses a very wide test benchmark. The database has a recognition rate of 99.15%, and the human eye is on the LFW. The recognition rate is 97.52%, and deep learning has exceeded the human eye recognition at the level of academic research.

Of course, when dealing with face recognition in real scenes, it is still unsatisfactory. For example, factors such as unclear face, lighting conditions, and partial occlusion will affect the recognition rate. Therefore, it is more appropriate to combine machine learning with manual confirmation in actual operation. There are many companies that do face recognition in China, including Shang Tang, contempt, and Yun Cong, Yi Tu, etc., which have deep data accumulation in the real environment or in the vertical segmentation field.

In addition to the explosive growth of facial recognition such as face and portrait, in addition. In the field of emotion recognition based on facial feature recognition technology, and gait recognition based on human body morphology, it is also rapidly developed on the basis of deep learning technology.

Speech recognition: Speech recognition has long been modeled using a mixed Gaussian model, and it has been a monopolistic modeling method for a long time, but although it reduces the error rate of speech recognition, it is still applied to commercial level applications. Difficulties, that is, in an environment where noise is actually not available. Until the advent of deep learning, the recognition error rate has dropped by more than 30% on the best basis of the past, reaching a level of commercial availability.

Due to the maturity of the speech recognition algorithm, the recognition rates of Keda Xunfei, Yunzhisheng, and Sibichi are similar on the universal recognition. In the promotion of the University of Science and Technology, the flight is the forerunner, from military to civilian, including mobile Internet, car networking, Smart homes are widely involved.

The maturity of the deep learning algorithm has greatly improved the accuracy of biometric identification, which ensures that the identification of human identity, including human face, form, mood, voice, etc., beyond the original scope, begins to move toward large-scale Application promotion. At the same time, advances in computer technology based on deep learning algorithms provide powerful computational and analytical tools for biometrics. In turn, the huge amount of biometric data also provides a wealth of material for machine training, "big data becomes the fuel of artificial intelligence."

Security application has broad prospects

At present, biometrics technology, which is widely used in security field, contains some people's identity characteristics based on computer vision. In addition to face and gait, it involves identity authentication, as well as fingerprints, irises, palm prints, veins, etc. These are all relying on computer vision technology to extract image patterns and identify content. There are also biometric recognition technologies such as speech recognition, brain wave recognition, and saliva extraction DNA.

In recent years, with the deep development of artificial intelligence deep learning and computer vision technology, several identity authentication technologies based on computer vision technology in biometric identification have also developed rapidly. In particular, face recognition technology, finger vein recognition technology, iris recognition technology and gait shape recognition technology are entering the golden age with the rapid development of big data, digitization and industry intelligence, and continue to combine the characteristics of industry segments. Deep application.

As the focus application in the security industry, biometrics technology is far superior to traditional passwords and credit cards in terms of technical security and market application. With the maturity of applications and the increasing awareness of consumers, the application of biometrics technology is rapidly becoming popular. The application of biometrics in the security business mainly includes the following categories:

The entrance and exit access control application is one of the most basic application areas of biometric identification technology. The earliest commercial application examples of biometric identification technology are almost all physical access control systems. In addition to the traditional credit card authentication method, fingerprint recognition is the most popular biometric recognition technology, and with the rapid decline in the cost of face recognition access control products and the continuous improvement of experience, it has a place in the middle and high-end market. For high-level security places such as banks and government agencies, the more secure finger vein and iris recognition technology has a larger application space. Overall, the share of access control products in the domestic biometrics market is second only to attendance products.

In addition, through the combination of biometrics technology and traditional electronic locks, it can actually be regarded as an extension of the application of physical protection products, such as fingerprint locks, face recognition locks, pupil recognition smart locks, vein recognition smart locks and so on. At present, along with the concept of “smart community” and “smart home” in China, such smart locks are showing rapid popularity.

The combination of biometric technology with the video surveillance system of “Safe City” and “Safe Community” is the most active performance of artificial intelligence deep learning. By embedding the face recognition algorithm or the human body gait algorithm and the facial emotion calculation in the video surveillance system, the personnel in the monitoring screen can be identified, or the warning is predicted, and various services can be realized through system linkage. Applications such as: when identified as a legal person, linkage open access control, personnel access, etc.; when identified as a blacklisted person, the linkage background generates an alarm prompt; through the personnel emotions and body shape is identified, the purpose of early warning can be achieved.

In this way, the application is no longer limited to the conventional after-the-fact forensics traceability, and gradually meets the new demands for personnel identity information acquisition and risk event early warning.

In addition to its application and development in the field of security, biometrics technology is rapidly expanding in people's daily lives. At present, fingerprint recognition has been popularized on smartphones. With the rise of full-screen mobile phones and the promotion of Samsung, Apple and other manufacturers, new in-screen fingerprint recognition, iris recognition, face recognition, voice recognition and other technologies will also become smart phones. New way to open. At the same time, biometrics are widely used in emerging fields such as smart homes and unmanned supermarkets.

Deep learning has helped to open up the “Diandu Ermai” in the field of biometrics, and will continue to approach 100% accuracy in the future, and the application field will be more extensive.

Editor in charge: Zhong Juanjuan