手指多模态生物特征粒化识别理论研究

Theorizing finger-based recognition is very significant for personal identification technology. However, it is very difficult to integrate the multimodal finger-features together using the current fusion principles. Viewing the finger trait as the combination of a fingerprint, a finger-vein and a finger-knuckle-print, in this project, we give a novel perspective to study the fundamental problems of multimodal finger-feature fusion according to some basic viewpoints of granular computing. First, a novel double-spectral polarized imaging system is proposed for jointly capturing fingerprint, finger-vein and finger-knuckle-print images. Second, based on the physiological structure of fingers and its imaging property in near infrared light, a novel method is proposed to effectively uniform the posture variations of fingers and interdependently locate the regions of interest of three imaging modalities. Third, in order to handle the feature fusion problem scientifically, a novel dynamic-granulating theory of multimodal finger-feature fusion is proposed by improving the traditional tolerance granulation space model. Finally, based on multiple granular computing, we propose a hierarchical and structuralized "stroke-model" framework of finger recognition. Hence, this project can provide sound theoretical principles for finger-based biometrics as well as promoting the cross integration between granular computing theory and biometrics, and the research findings will be very significant for the development of biometric technology.

以手指整体为研究对象探索人的身份鉴别方法具有重要的科学价值，但受当前融合识别理论的制约，实现手指多模态生物特征的有机融合是十分困难的。本项目视指纹、指静脉和指节纹为构成手指生物特征的三个基本单元，根据粒计算的基本观点，从一个新的理论视角探讨手指多模态生物特征融合与识别所面临的一些基本问题。首先，根据偏振光成像原理，提出了一个双光谱手指多模态生物特征图像综合采集新方案；其次，根据手指生理结构和成像特点，提出了手指多模态特征之间的姿态统一和协调定位新方法；然后，为突破困扰特征融合的瓶颈，以传统相容粒度空间为基础，提出了手指多模态生物特征动态粒化融合新理论；最后，以多粒度计算为基础，提出了层次化和结构化的"笔画型"手指生物特征识别新框架。本项目的研究不仅能够解决手指多模态生物特征融合与识别方面存在的基本科学问题，而且还能有效促进粒计算与生物识别理论的交叉与融合，其研究成果具有重要的理论价值。

本项目基本解决了项目计划书所设定的各项研究问题。在手指多模态成像模式上实现了三种成像装置和一个嵌入式手指识别系统，为深入认识手指多模态识别问题奠定了很好的基础。通过标定和空间映射，实现了手指多模态特征图像的姿态矫正和ROI协调区域定位。通过三种粒化方法，验证了相容粒度空间在手指多模态融合识别方面的性能，并发现了它的一些固有缺点。针对传统方法的不足，提出了适合手指三模态粒化融合的超球粒度空间模型，并建立了符合“笔画型”识别理念的多粒度、层次化的超球商空间模型，此模型在提高识别精度的同时，还能大大降低计算代价。依托本项目，针对手指静脉识别的其它问题，如静脉方向场提取、图模型结构、手指签名生成、静脉血管的深度分割和静脉网络复原等，项目组也做了成功的探索，并取得了很好的结果。项目的研究为我们深入理解手指多模态识别中的关键问题和掌握关键技术提供了有力支撑，为开发切合实际的手指识别系统奠定了良好基础。