隨著科技的日新月異,傳統使用來作為安全控管及身分辨識的方式已逐漸不符合時代的需求。因此近代發展出利用生物特徵作為系統管理的識別工具,且已成為一種方便又有效的安全控管方式,這其中又以指紋驗證最具有實用性與代表性。 本論文所研究的是有別於光學式的熱感式指紋。在熱感式指紋辨識技術核心演算法上,本論文主要是探討指紋重建、登入與特徵比對等三大部分。在指紋重建方面,主要是以最小絕對平均誤差求出指紋片段之間的關係性,以達到消除冗餘性之目的,並討論與分析不同的堆疊方式所造成之影響﹔在登入方面,本論文提出以片段與片段之間的冗餘程度訂出登入準則的概念;在特徵比對方面,除了嘗試核心點搜尋演算法以及結構比對演算法外,還提出一個改良式結構比對演算法的架構。 從實驗結果中可獲知,重建出的影像在清晰度與穩定度上皆優於從熱感式伴隨展示之軟體中所擷取的指紋影像;本論文所提出之改良式架構,在方向角不準確或者指紋影像整體縮放的情況下,皆能有效的辨識成功。 關鍵字:生物特徵、認證、指紋辨識、重建、特徵比對、冗餘性
With the rapid development of science and technology, the traditional method for security management and personal identification is gradually obsolete. Therefore, modern biometrics has been developed as an identification tool for system management, and it has become a convenient and effective security control measure. Among all possible ways of biometrics, fingerprint verification is the most practical and representative. The thesis studies the thermal instead of optical fingerprints. In the kernel algorithm of thermal fingerprint identification techniques, this thesis will focus on the study of fingerprint image reconstruction, enrollment and minutiae matching. For the fingerprint reconstruction, the minimum absolute average error is applied to exploit the correlation between fingerprint slices so that the redundancy among slices can be removed. In addition, different slice stacking methods and their effects are discussed and analyzed. For the enrollment, the idea of establishing the enrollment criterion according to the degree of redundancy between slices is proposed. For the minutiae matching, we not only investigate the core finding and the structural matching algorithms, but also present an improved framework of the structural matching algorithm. In terms of clarity and stability, images reconstructed are superior to those obtained by the demonstration software associated with the thermal sensor. The improved framework can effectively identify the fingerprint even when the orientation is not correct or the fingerprint image is scaled globally. Keywords: Biometrics, Verification, Fingerprint identification, Reconstruction, Minutiae matching, Redundancy