Using ECG as Biometric Trait: Approaches based on Variant CNN Architectures

One of the latest advances in the field of biometric human identification is the use of Electrocardiogram (ECG). Approaches including time domain, frequency domain and recently, deep learning methods have been implemented to carry out this task. But, results show that, there is much room for improvement. This paper proposes two different variant Convolutional Neural Network (CNN) architectures, based on residual learning and dense learning schemes. The aim is to implement the best network architecture to classify /identify ECG waveforms based on four criteria namely, simplicity of architecture, number of patterns required to be trained for best results, accuracy obtained and training time required. The potential of the models have been tested on a dataset of 60 subjects collected from four public databases. The performance of the proposed models have also been cross checked with two other deep learning approaches namely, Long Short Term Memory (LSTM) network and transfer learning using Resnet-50. Profuse investigation under the given conditions shows that, the proposed dense learning network is the winner.