In this project, we utilize the electrocardiogram (ECG) as a primary biometric modality in human identification. The design steps of the proposed approach are the following: first, we segment the ECG signal and utilize its cyclostationarity and spectral correlation to enrich the signal’s original informational content. Then, we generate spectral correlation images. During this process, we disregard the time consuming algorithmic step, typically used in other similar ECG based machine learning approaches, namely the fiducial point’s detection and noise removal steps. Next, our spectral correlation images are fed into two convolutional neural networks (CNN) architectures, which we fine tune, test and evaluate, before we suggest a final architecture that demonstrates improved ECG based human identification accuracy. To evaluate the efficiency of the proposed approach, we perform cross validation on nine, small and large scale, ECG databases that encompass both normal and abnormal ECG signals. Experimental results show that independent of the database used, our approach results in improved system performance (compared to state of art approaches), yielding an identification accuracy, false acceptance and false rejection rates of 95.6%, 0.2%, and 0.1% respectively. This project is implemented with MATLAB software.