生物特征识别学科发展报告

From mobile phone unlocking and community access control to dining in restaurants, supermarket cashiering, high-speed rail station entry, airport security checks and hospital visits, biological characteristics such as face, iris and fingerprint have become the digital identity cards for people to enter the Internet of Everything world. Biometric recognition endows machines with advanced intelligence to automatically detect, capture, process, analyze and recognize digital physiological or behavioral signals. It is a typical and complex pattern recognition problem and has always been at the forefront of the development of artificial intelligence technology, occupying an important position in national strategies such as the new generation of artificial intelligence planning and the "Internet +" action plan. Due to the issues of privacy, ethics and law that are of vital concern to the public involved in biometric recognition, it has also attracted extensive social attention recently. This article systematically reviews the current development status, emerging directions, existing problems and feasible ideas of the biometric recognition discipline, and thoroughly combs the research progress of face, iris, fingerprint, palm print, vein, voiceprint, gait, person re-identification and multimodal fusion recognition. Taking the face as an example, it focuses on introducing the new directions that have received attention in the field of biometric recognition in recent years - adversarial attacks and defenses, deep fakes and anti-fakes. Finally, it analyzes and summarizes the three major challenging problems existing in the field of biometric recognition - "perception blind area", "decision-making误区" and "safety red zone". This article believes that it is necessary to transform and innovate the sensing, cognitive and security mechanisms of biological characteristics in order to possibly achieve fundamental breakthroughs in academic research and technical applications of biometric recognition in complex scenarios, break through the drawbacks of existing biometric recognition technologies, and develop towards the overall goal of the new generation of biometric recognition of being "sensible", "knowable" and "trustworthy".

从手机解锁、小区门禁到餐厅吃饭、超市收银,再到高铁进站、机场安检以及医院看病,人脸、虹膜和指纹等生物特征已成为人们进入万物互联世界的数字身份证。生物特征识别赋予机器自动探测、捕获、处理、分析和识别数字化生理或行为信号的高级智能,是一个典型而又复杂的模式识别问题,一直处于人工智能技术发展前沿,在新一代人工智能规划、“互联网+”行动计划等国家战略中具有重要地位。由于生物特征识别涉及公众利益攸关的隐私、道德和法律等问题,近期也引起了广泛的社会关注。本文系统综述了生物特征识别学科发展现状、新兴方向、存在问题和可行思路,深入梳理了人脸、虹膜、指纹、掌纹、静脉、声纹、步态、行人重识别以及多模态融合识别的研究进展,以人脸为例重点介绍了生物特征识别领域近些年受到关注的新方向———对抗攻击和防御、深度伪造和反伪造,最后剖析总结了生物特征识别领域存在的3大挑战问题———“感知盲区”、“决策误区”和“安全红区”。本文认为必须变革和创新生物特征的传感、认知和安全机制,才有可能取得复杂场景生物识别学术研究和技术应用的根本性突破,破除现有生物识别技术的弊端,朝着“可感”、“可知”和“可信”的新一代生物特征识别总体目标发展。