In recent years, electroencephalography (EEG) signals have gained attention for biometric applications. The nature of EEG-based identification involves the detection of distinct patterns from complex spatio-temporal signals. In this study, we aim to investigate the effect of transfer learning and the signal enhancement on permanence and distinctiveness. Two network models, the deep convolutional neural network and the autoencoder, are used for automated feature map generator. For signal enhancement, we utilized the Laplacian of Gaussian and proposed a Softmin operator. We explored the effectiveness of two stimuli, the resting-state with eye-open and closed. We extracted several statistical features over epochs. Initial results show that the transfer learning-based feature representation and signal enhancement are promising for EEG-based biometric applications.