基於心電向量圖的強健性身份辨識系統

Abstract

心電圖具有高度的個體間差異性，是近年來研究聚焦的生物識別特徵。前人研究大多選擇單一肢導程作為訊號來源，其正確率往往受限於受測者的生理及心理狀態。為了提升身份辨識的強健性，本論文提出了一個以多導程心電圖為基礎的身份辨識系統。首先將一般心電圖儀量測的三個標準肢導程透過矩陣轉換，產生由Frank Lead X、Y與Z展開的三維度衍生心電向量圖。再配合不同的特徵參數擷取機制，使用以支持向量機為基礎的分類器進行身份識別。至於特徵參數擷取機制，我們分別探討傅立葉描述子與主成分分析參數的身份辨識性能。有別於傅立葉描述子由平面投影取得的二維資訊，主成分分析參數可具體描述心電向量圖的三維軌跡。實驗結果顯示，多導程心電圖的整合確實有助於提升身份辨識的強健效能。針對PTB Diagnostic資料庫的50名正常人與123名心肌梗塞患者，使用一個心跳週期的辨識率分別可達99.98%與99.22%。

In recent years, ECG has become a popular biometric identification feature. Unlike previous works based on one-lead ECG, this work aims at improved robustness by incorporating multi-lead ECG in biometric identification. Firstly, the three standard limb leads are transformed by a conversion matrix to generate a three-dimensional(3D) derived vectorcardiogram (dVCG), which is composed of the three Frank leads, X, Y and Z. Two types of biometric features are investigated. Fourier descriptors are extracted from the X-Y planar projection of dVCG, whereas principal component analysis (PCA) coefficients can describe more closely the 3D trajectory embedded in the dVCG. Together with the feature extraction, a classifier based on the support vector machine is applied to achieve person identification. Experiments on PTB diagnostic database demonstrate the validity of the proposed PCA method with identification accuracies of 99.98% and 99.22% for 50 healthy subjects and 123 myocardial infarction patients, respectively.