指纹方向场高维数据分析方法及其应用研究

Biometric fingerprint technology has a wide range of vital applications for protection of personal and national security. Despite great commercial success, there are still significant challenges with the technology. Today, manual intervention is still indispensable in latent fingerprint identification and spoofing attack is still one of the major security threats to automatic fingerprint systems. Ridge orientation estimation plays a central role in the entire fingerprint feature extraction process, and hence is a key to solving these problems. To advance the technology, it is important to understand the fingerprint feature, its principal characteristics from data as well as patterns and sources of variation in the context of specific applications. Such a study will require statistical analysis of ridge orientation field, which is a challenge due to the curse of dimensionality using conventional methods. This project aims to tackle the problem by modeling samples of ridge orientation field as functional data. This novel interpretation will enable us to systematically address three closely related issues in a unified and theoretically sound framework: 1) representing the conventional discrete data of ridge orientation field as functions, 2) studying important sources of pattern and variation among data using functional representations, and 3) applying them to latent fingerprint identification and anti-spoofing techniques. The project is placed at the cutting-edge interface of modern statistical methodologies and biometric security applications. The expected outcome will help to advance application and study of the theory of functional data analysis, understand important patterns and modes of variation in ridge orientation features, and accordingly, improve the reliability and privacy measures of biometric security systems. Therefore, the project has both theoretical and practical values, and has the potential to generate a promising impact in both academic and commercial terms.

自动指纹识别系统应用广泛，对人们生活、国家和公共安全都有深刻影响。但目前技术还面临现场指纹识别依赖人工辅助、电子欺骗威胁自动指纹识别系统安全等问题。脊线方向场估计是整个指纹特征提取算法的核心，也是解决上述难题的关键之一。相关技术的进一步提高需要对大量方向场数据进行分析，挖掘脊线形态主要模式及其变化关联的一般规律。但指纹方向场的传统高维数据表示方式对特征分析造成了困难。本项目将结合指纹识别研究经验和先进的高维数据分析方法，将方向场离散数据转化为函数表示，然后利用函数特性来研究方向场特征变化关联的典型模式及其在现场指纹残缺方向场重建和真、假指纹形变模型中的应用。本项目立足于生物安全和统计分析的跨学科领域，其顺利开展将有助于推动相关基础理论的应用研究和拓展，为自动指纹识别系统的安全性和可靠性提供有力的理论依据和技术支持，帮助充分实现指纹生物特征作为身体密码和加强公共安全方面的价值。

本项目通过构建数理模型分析生物特征样本个体类型和样本差异的变化特征掌握和运用生物特征内部关联变化的一般规律，并以此作为先验信息提高生物特征识别系统的自动处理能力与识别精度，进而为提升生物特征识别系统的安全性和可靠性提供有力的理论依据和技术支持。从本项目提出的指纹相似性检索问题出发，我们通过对高维特征的数据分布建模分析，对包括指纹识别在内的生物特征查询如何同时保障系统准确性、高效性、安全性及可靠性进行研究。主要研究内容包括指纹特征空间分布统计模型、高维指纹特征的紧凑二进制码表示、特征函数分析在指纹检索中的应用以及生物特征查询系统的隐私保护机制。作为项目主要研究成果，我们发表了2篇计算机学会推荐A类国际学术类期刊论文，满足资助项目推动相关基础理论研究与应用的要求，基本完成了项目设定的帮助提高生物特征作为身体密码和加强公共安全方面价值的目标，有利于进一步增强中国生物特征识别产业的技术水平和竞争力。