Privacy-Protected Human Identification in Encrypted/Transformed Domains

在过去的二十年里，生物识别技术在许多领域得到了广泛的应用，如医疗保健、银行、监控和安全控制。鉴于全球互联网和移动计算的日益普及，许多公司已转向生物识别隐私和安全，以确保安全通信。然而，第三方或公共网络服务器上的生物特征验证可能会以未经授权的方式被滥用。为了保护隐私，提高安全性，人们提倡在加密或变换的域中进行生物特征验证，这样可以更有效地保证隐私和安全。该项目背后的基本思想是，不可逆加密/变换域中的生物特征包含的信息量与其原始信息量完全相同，因此可以建立一种模式识别方法来从加密/变换域中的混沌信号中确定/提取有用信息。这是第一个Grant方案项目，旨在研究如何从混沌信号中发现和评估信息以获得辨别能力，并开发用于加密/变换域中的生物特征/多生物特征验证的健壮模式识别方案。所提出的方法/方案将在典型的野生人脸/语音/步态数据集上进行强有力的验证，并将设计并在真实环境中测试两个实际演示系统(生物特征银行和行人特征识别)。该项目将专注于对混沌信息的理论理解和针对特定应用的混沌模式识别的开发。考虑到隐藏在一组给定的混沌信号下的多个数据结构，我将开发一种稳健的方法来找出潜在的各种数据结构，用于数据理解、聚类和分类。另一方面，考虑到加密/变换生物特征验证等特定问题，需要研究这一特定主题中的一般理论结果，并开发一个稳健的生物特征身份识别方案。该项目的工作涉及信号处理、机器学习和模式分析等领域。加密/变换生物特征验证的研究源于英国新兴新业务的实际新需求。该项目将提供必要的理解，以便未来开发具有新应用的健壮的生物测定验证方法。

Biometrics has been widely utilized in the past two decades in many areas such as healthcare, banking, surveillance, and security control. Given the increased uptake of internet and mobile computing globally, many companies have been turning to biometric privacy and security to ensure secure communication. However, biometric verification over third-party or public network servers may be abusively exploited in an unauthorized way. To protect the privacy and improve the security, it has been advocated to carry out biometric verification in encrypted or transformed domains, where privacy and security can be more effectively guaranteed. The basic idea behind the project is that the biometrics in the irreversible encrypted/transformed domains contains exactly the same amount of information as its original one, and hence one can establish a pattern recognition methodology to determine/extract useful information from chaotic signals in encrypted/transformed domains. This First Grant Scheme project aims to investigate how to discover and evaluate the information from chaotic signals for discriminative power, and develop robust pattern recognition schemes for biometric/multi-biometric verification in encrypted/transformed domains. The proposed methods/schemes will be vigorously validated over typical wild face/speech/gait datasets, and two practical demo systems (biometric banking and pedestrian profiling) will be designed and tested in real world environments.The project will focus on both theoretical understanding of chaotic information and application-specific exploitation of chaotic pattern recognition. Considering multiple data structures hidden beneath a set of given chaotic signals, I will develop a robust way to find out the underlying various data structures for data understanding, clustering and classification. On the other side, given a specific issue such as encrypted/transformed biometric verification, one need to examine the generic theoretic findings in this specific topic and develop a robust scheme for biometric human identification.The work of this project is within the areas of signal processing, machine learning and pattern analysis. The research on encryted/transformed biometric verification has come from the practical new needs of the UK's emerging new businesses. The project will provide the understanding needed to allow the future development of robust biometric verification methods with novel applications.