Robust Pupil Segmentation using UNET and Morphological Image Processing

The current development in image processing towards biometrics systems has opened much research on realtime applications. The deep learning algorithms are added many expectations to the researchers. The main challenges of these applications are vulnerability towards training time, detection accuracy, and accurate segmentation. In addition to this, the visual noise among various biometric systems is the main challenge. In this paper, we deployed the CNN model using modified UNet to perform the segmentation. The proposed method uses noisy images from the MMU (Multi Media University Iris database) dataset. The acquired colored eye images from the dataset exhibit specular reflections, eye gaze, off-angle images with less resolution, and occlusions caused by eyelids and eyelashes. The focus of our work is mainly to perform accurate segmentation in less training time. Compared the existing methods that uses UNet architecture, with the proposed method, we achieved an accuracy of 91.7%.