Parallelization Of ECG-Based Biometric Recognition

During last decade the use of ECG recordings in biometric recognition studies has increased. ECG characteristics made it suitable for subject identification: it is unique, present in all living individuals, and hard to forge. These characteristics make it suitable for objects' identification. Nowadays, a vast of ECG signals has been used into Biometric AuthenticationWith the big data showing up, the traditional ECG feature analysis and pattern matching algorithms such as SVM, LDA, LSH calculation mode, etc, are accused of the high time-cost when facing mass data processing. Indeed, all types of pattern matching algorithms, the parallelization of mass tasks and throughout rate has always been the bottleneck of the whole system.Therefore, based on the fiducial-point and Non-fiducial approaches, this paper proposes a combination of Non-fidicuial and fiducial mixing characteristics, the mixing features include: fiducial position ECG signal cycle; ECG morphology cycle. And based on ECG morphology cycle modeling, we propose the mixed ECG signal characteristics.The second stage, this paper based on the ECG fixed features, proposes the advanced LDA algorithm based on Multiple Features(LOMF), utilizing Map/Reduce distributed computing framework, accomplishing the ECG signal preprocessing, Brisk feature extraction and two-level hash retrieval, improving the computing efficiency while using LDA between substring classification, recognition accuracy was dragged into a higher level. In response to large-scale data analysis, LOMF algorithm support incremental training, and enhance the recognition rate of the system to a maximum extent.The paper comparing the discrepancy beween the mixed ECG features and fidicuial features, Non- fidicuial features, it was found wehn working under the same kind of recognition algorithms, ECG mixed feature shows higher retrieval accuracy. And based on sub-string theory the improved LOMF shows more 7% to 8% over the traditional algorithms, and LOMF biggest advantage is supporting incremental training, fit in with the distributed system(Map/Reduce)framework, coming more suitable for the Internet with rapid growth data.