Insight | Federal learning, formation encryption, fuzzy extractor? A new generation of biological identification technology enhanced privacy protection

"Zhi Zhou Insight" is a new research brand of machinery PRO, focusing on new generation of new generation of artificial intelligence technology, surrounding artificial intelligence academic exploration and engineering technology hot research direction in-depth exploration, tracking the latest AI technology route, analyzing mature Situation, insight into potential technology iterative opportunities. "Zhou Yaojing · Trusted AI" special series focuses on building a new generation of artificial intelligence technology for the purpose of building credible AI systems, expanding specific discussion around privacy protection, interpretable, fairness and robustness increase.

Write | Machine heart Pro

This section is selected from the "Zhou Insight" report series "Available Ai" topic "Privacy Protection Enhancement New Generation Biological Recognition Technology". The core of the machine will continue to surround a new generation of AI technology to build "credible AI" to issue technical insights, survey reports, organized online sharing and technical discussions, etc. Get full verse reports and event notifications.

This report has received many experts with great professional opinions. Special acknowledgment (sorted by surname’s first letters): Harbin Institute of Technology Science and Technology, Ph.D. Han Jiqing, Dr. Lei Zhen, Dr. Lei Zhen, Zhejiang University Network Professor Renkui, Dean of Space Safety, Dr. Tang Linpeng, Chief Technology Officer of Muqi Technology, Professor, Department of Computer Science and Engineering, Shanghai Jiaotong University, Professor Zhang Junping, Computer Science and Technology, Fudan University


  • Dual-law implementation, enhance the privacy protection capacity of biometric system to reduce system security risk, abuse tracking and governance technology, biological information deducting, anonymization, etc.

  • AI Times Biological identification application broke out, biometric authentication replaces password in a large number of scenes, discloses, and the biological template cannot be changed or recovered once it is leaked. Enhanced the protection of biological templates is a key link to enhance biometric system privacy protection capabilities

  • The new generation of credible biometric systems should have irreversible, revocable, non-associated technical features. Feature transformation, biological encryption can enhance protection of biological templates.

  • Head Internet enterprise, AI technology company laboratory, research team has launched the privacy protection technology upgrade of biological identification system, such as strengthening biological template encryption technology research, using federal learning, multi-mode fusion and other computational and verification methods; Some emerging companies are also exploring a biometric technology solution that is more easy to use, higher security, and privacy privacy.

With the implementation of the data security law, the implementation of the personal information protection law, the market is upgraded to the upgrade technology, using the new generation of privacy protection to improve the need to heat up the needs of the biometric system, from the technical breakthrough, improve the safety of biometric system, and realize biometric identification The privacy capacity of the system is imminent. Accordingly, the research and development of biometric technology to construct a new generation of confident AI system capabilities is also rising.

In October 2020, China Xinyiyuan Terminal Laboratory United Telecom Terminal Industry Association, Telecom Terminal Industry Association, China Internet Association, Internet Financial Identification Alliance and Mobile APP Innovation and Personal Information Protection Laboratory joint release "Biological identification privacy protection The research report ", eight dimensions for the privacy protection of biometric system, and extended the technical topics that need to be solved, the machine’s heart is summarized as follows:

New generation biometric technology design for privacy protection

When designing a biometric identity authentication system architecture, the full process of biological data security and biometric verification accuracy is the two most important aspects that need to be considered simultaneously [6]. At the time of the technical selection of biometric applications, the recognition accuracy and ease of use of different biometric techniques also have a large difference, and the following figure shows the different types of biometric technology in terms of recognition accuracy and ease of use. Case.

The credible AI system should have the ability to protect the biological information template when using biometric identification. The ISO / IEC 30136 standard is given a method of evaluating the accuracy, confidentiality and privacy of the biological identification template, specifically indicating that in terms of security and privacy protection, IRREVERSIBILITY, unlinkability Features.

National Standard "Information Technology Safety Technology Biological Characteristics Identification Information" (Draft for Comment) also pointed out that biometric identification systems should meet confidentiality, integrity, and updateability and cancellation, and specify biometric identification information. The following requirements: irreversibility, negligibility, confidentiality. Furthermore, the available biological template protection mechanism achieves the objective features such as irreversibility, cancellation, non-correlation, and features of Performance (Performance) [9].

Increase the core fortress of the privacy capacity of biological identification system: biological template protection

Biometric identity certification is generally divided into two stages of registration and verification. The common biometric identity authentication system architecture is as follows, where t represents the biometric sample obtained when registration, Q is the query biometric sample obtained when identified, XT and XQ are template and query feature set, S represent match results [6 ].

The biometric data is scanned in the registration phase to determine a group of different biometric sets, called biometrics templates. Bioinformatics template data leaks are also the focus of personal information protection. Unlike passwords or secret key schemes, the biological information template is unlike password, once the leaks cannot be changed or recovered [8]. The following figure shows the main security vulnerability zone in the biometric system [6].

Biological template protection: technical methods such as characteristic transformation and bioconagration

Protecting biometric templates from a variety of hacker attacks is a vital issue. Protecting biometric templates Mainly include characteristic transform and biological encryption two sets of technical routes [6], including template encryption, can revoke biological identification, mixing methods, formative encryption, visual cryptography, etc. [6] The following table further summarizes the relevant technical methods.

Trusted biological identification technology upgrade: large factory and start-up enterprises have joined racing

Head Internet enterprise, AI technology company laboratory, research team has launched the privacy protection technology upgrade of biological identification system, such as strengthening biological template encryption technology research, using federal learning, multi-mode fusion and other computational and verification methods; Some emerging companies are also exploring more easy, high security, privacy biological identification programs and products, such as Miqi Technology, with non-contact finger and refer to the vein collection method, using multi-scale-based multi-scale Feature extraction and heterogeneous high performance ratio algorithm, push privacy protection enhanced biometric identification in a multi-scale image representation and cryptographic binding.