本論文研究了指靜脈識別系統，該系統採用了一種安全的生物特徵識別方法，由於其固有的防偽性和非侵入式的獲取過程，使其成為可靠的選擇。論文引入了一種優化的指靜脈識別方法，該方法利用深度學習，著重於抗鋸齒技術和注意力機制的整合，以提高識別精度。 研究分為三個主要部分。首先，我們開發了一種針對指靜脈識別的卷積神經網絡（CNN）架構，其中包含了旨在減少失真並保留靜脈圖案細節的新型抗鋸齒濾波器。其次，我們整合了注意力機制，特別是通道和空間注意力，這些機制選擇性地增強了對於靜脈圖案辨識至關重要的特徵。這種雙重方法同時解決了由於鋸齒現象而造成的細節損失和在復雜圖像中突出相關特徵的挑戰。 在多個基準數據集上進行的廣泛實驗證明了所提出方法相比傳統指靜脈識別系統的優越性。我們的結果顯示了在識別精度和對各種圖像質量問題的魯棒性方面的改進，確認了在深度學習模型中結合抗鋸齒技術和注意力機制的有效性。本工作不僅推進了指靜脈識別技術的發展，還提供了可能適用於其他生物安全領域的洞見。 ;This thesis is a study of the finger vein authentication system, utilizing a secure biometric method due to its intrinsic resistance to forgery and its non-intrusive acquisition process. It introduces an optimized approach for finger vein authentication using deep learning, focusing on the integration of anti-aliasing techniques and attention mechanisms to enhance authentication accuracy. The research is divided into three main parts. First, we develop a convolutional neural network (CNN) architecture tailored for finger vein recognition, incorporating novel anti-aliasing filters designed to mitigate distortion and preserve vein pattern details. Second, we integrate attention mechanisms, specifically channel and spatial attention, which selectively enhance features relevant for vein pattern discrimination. This dual approach addresses both the loss of detail due to aliasing and the challenge of emphasizing relevant features in complex images. Extensive experiments conducted on several benchmark datasets demonstrate the superiority of the proposed method over traditional finger vein authentication systems. Our results show improvements in recognition accuracy and robustness against various image quality issues, confirming the effectiveness of combining anti-aliasing techniques and attention mechanisms in deep learning models. This work not only advances the state of finger vein authentication but also offers insights that could be applied to other areas of biometric security.
