基于判别特征学习的非接触式掌纹识别方法研究

Palmprint contains rich highly discriminative features that are able to uniquely identify an individual. Palmprint-based biometrics can be widely used in the security-check, access-control，attendance-checking and forensic applications. Recently, contactless palmprint images, which are captured under unconstrained conditions, show high user-friendliness and applied-convenience. Therefore, contactless palmprint recognition technology should be one of the most promising research direction of palmprint recognition. However, contactless palmprint images are possibly variant on positions, translations, violations, opening-degree and illuminations, which result in some difficulty of feature extraction and relatively low accuracy of palmprint recognition using conventional representative methods. Therefore, how to extract the discriminative features of contactless palmprint images remains a central and challenging problem. In this research project, we have an aim to study the discriminant feature learning and recognition of contactless palmprint images for more accurate palmprint recognition: 1) We comprehensively analysis the characters of contactless palmprint images, and design a model to calculate the representation ability of palmprint orientation features, based on which we propose to design an optimal orientation feature descriptor to represent contactless palmprint images; 2) We will define a new distance measure criterion for orientation feature descriptor of contactless palmprint images, and further propose linear discriminant orientation features learning and low-rank discriminant orientation features learning models for contactless palmprint recognition; 3) Furthermore, we will study the discriminant feature learning and recognition of the video-based contactless palmprint images based on discriminant orientation features learning. If this study can be carried out successfully, it will be helpful to promote the real-world application of contactless palmprint recognition, and further it will improve and enrich the basic theory of the contactless palmprint recognition.

掌纹包含丰富的身份判别信息，基于掌纹的生物特征识别技术在安检、门禁、考勤和刑侦等领域具有广泛的应用前景。非接触式掌纹识别是目前最具应用价值和应用便捷性的掌纹识别技术。然而，非接触式掌纹图像因受手掌摆放方式变化以及外界环境等因素影响导致其特征提取难、识别率偏低，这极大阻碍了掌纹识别技术应用的推广。本课题专注于如下非接触式掌纹判别特征提取与识别问题：1)基于掌纹方向特征表示能力，研究非接触式掌纹图像最优方向特征表示；2)基于非接触式掌纹最优方向特征表示，研究掌纹方向特征距离度量问题，并重点研究非接触式掌纹图像线性判别方向特征学习以及低秩判别方向特征学习理论和识别方法，为本质上提高非接触式掌纹识别精度提供理论依据；3)将判别方向特征学习理论进一步扩展到动态非接触式掌纹图像判别特征提取和识别。本项目的研究成果，将有助于推广非接触式掌纹识别实际应用，并进一步完善和丰富非接触式掌纹识别的基础理论。

生物特征识别是信息时代身份认证的重要技术手段。作为一种新兴且非侵犯的生物特征技术，掌纹识别技术在门禁、安检、考勤以及移动身份认证等民用和商用领域具有广泛应用前景。本项目围绕非接触式掌纹识别，研究非接触式掌纹图像ROI提取预处理和掌纹最优判别特征提取与表征，并重点研究基于判别特征学习的非接触式掌纹识别方法。.经过为期三年的研究，本项目在非接触式掌纹识别方面取得较好的研究成果，主要包括以下四个方面：1）全面系统分析当前掌纹特征提取与识别方法，撰写掌纹识别技术综述性论文；2）设计一种非接触式掌纹图像ROI提取策略；3）提出一套掌纹方向特征表示，进而设计了一种掌纹方向特征与其判别能力计算模型，为掌纹方向特征选择提供了理论依据，并基于此提出一种自适应最优判别方向特征提取与表示方法，较传统掌纹方向特征表示具有更优的识别性能；4）提出一种掌纹方向判别方向特征学习模型与掌纹识别方法，进而提出一种基于方向和纹理多类型特征联合学习的掌纹识别方法，有效地提升了掌纹识别性能。此外，本项目还在3D掌纹识别方面进行了初步研究并取得一定的研究成果。本项目相关成果累计发表学术论文18篇，包括期刊论文13篇和会议论文5篇。其中，IEEE Transactions期刊论文9篇，CCF A类论文2篇，CCF B类论文8篇和CCF C类论文1篇，以及“ESI高被引”论文1篇。.本项目的研究具有较强的应用价值和科学意义。从应用的角度，本项目提出的掌纹最优方向特征表征与识别方法可为身份识别提供一种新的便捷且非侵犯的身份识别思路，如在门禁、安检以及以移动支付等领域，人们可以很方便地通过“刷手”完成信息采集和身份认证。从科学发展的角度，传统基于单一生物特征的识别系统在实际应用中不可避免地存在一定的局限性，发展多类型生物特征识别是生物特征识别发展的主要趋势，本项目在掌纹识别方面的相关研究进一步完善了生物特征识别理论体系，为智能互联网时代的生物特征识别应用与发展探索新思路。