基於分散式訊源編碼架構之心電圖身份認證系統

Abstract

傳統心電圖認證系統仍有個資外洩之疑慮，本論文將認證系統建構於分散式訊源編碼架構之上以增強心電圖身份認證系統之安全性能。個人獨特資訊的擷取，是將單一導程的一維心電圖訊號轉變成二維心電圖影像，藉由JPEG2000壓縮與部份解壓縮機制取出特徵參數，再配合量化與位元篩選程序，最後使用LDPC碼進行解碼。利用位元錯誤率分布的差異，LDPC碼能夠將使用者有效區隔為合法或非法侵入兩類。實驗測試資料由QT與Fantasia資料庫取得，包含心臟正功能常與心律不整病患。系統模擬探討各種特徵集合、心電圖影像與心臟健康狀態對於認證效能的影響。實驗結果顯示心電圖認證系統能夠彈性調整LDPC碼之更正能力以符合不同層級的安全性需求，對於心臟功能正常者的認證效能達到 ，加入心臟病患則夠達到 的認證效能。

In recent years, Electrocardiogram (ECG) has been proposed as a new biometric modality for human authentication and access control. ECGs are often compressed for efficient transmission and storage purposes and therefore, incorporation of compressed domain ECG biometric is preferred as it by-passes the full decompression. To further secure the biometric data, we present a new biometric storage system using low-density parity-check (LDPC) codes in distributed source coding (DSC) framework. We start with the direct feature extraction from the compressed ECG images in JPEG2000 format. Biometric features are then subject to DSC encoding to compute syndrome bits, which have certain advantages over original features in terms of security and memory size. When combined with the side information at the DSC decoder, the syndrome bits provides robust biometric authentication for genuine users and guards against attacks from imposers. Experiments on the QT and Fantasia databases demonstrate the validity of the proposed system with high accuracies for human authentication.