IPSegNet: Deep convolutional neural network based segmentation framework for iris and pupil

In biometric based authentication system, Iris is one of the most extensively used biometric trait as it has seen ground breaking research in both region of interest extraction and recognition. Several researchers in the field of biometric based authentication systems have claimed that the main reason for several matching errors is the poor segmentation of the trait. The task of segmentation for a biometric based authentication system is one of the most crucial, as most of the matching and recognition algorithms are performed on the particular region of interest in an acquired image. In this paper we propose two novel end to end convolutional neural network based architectures for region of interest extraction in an iris. The proposed architectures take the image of an eye as input and produce two circular regions of interests for the Iris and Pupil respectively. The two networks proposed were inspired from two state of the art object detection networks Faster RCNN (Region based Convolutional Neural Network) and SSD (Single Shot Multi-Box Detector), both these modified networks were trained for 8000 images. Experimental analysis performed proves that both of our techniques have very high performance in terms of accuracy's and overlaps.