Fast and robust retinal biometric key generation using deep neural nets

For biometric identification with retina, vascular structure based features bear significance in preparing retinal digital templates. This requires analysis of large amount of data from different sources. It needs a faster and robust automated system to extract the quantitative measures from huge amount of retinal images. Therefore, fast and accurate detection of existing retinal features is an important element for a successful biometric identification. In this work, we propose to design and implement a retinal biometric key generation framework with deep neural network. The purpose is to replace the semi-automated or automated retinal vascular feature identification methods. The approach begins with segmentation from coloured fundus images, followed by selection of some unique features like center of optic disc, macula center and distinct bifurcation points on a convolutional neural network model. For better understanding, the key generation process has finally been shown with the help of a graphical user interface. This network was trained and tested with the training and testing images of DRIVE dataset and some of our previously published result sets on automated feature extraction methods. The network was trained on NVIDIA Titan Xp GPU provided by NVIDIA corporation.