Paper Title
Multi-Region Eye Segmentation Using Attention Gates and Deep Convolutional Neural Networks

Abstract
Accurate segmentation of eye images into individual regions, i.e., the background, sclera, iris, and pupil, is highly critical for medical diagnosis, biometric authentication, and human-computer interaction. Here, we propose a novel attention-based deep convolutional neural network (CNN) model for multi-region eye segmentation. The proposed method utilizes attention gates to enhance feature selection and improve segmentation accuracy. Its performance is evaluated using the SBVPI and UTIRIS datasets and compared to the state-of-the-art segmentation methods, including U-Net and SegNet. Experimental results reveal that the proposed model achieves Dice Coefficient (DC) value of 0.9320, Jaccard Index (JI) of 0.8755, precision of 0.8948, recall of 0.9770, and accuracy of 0.9778, better than U-Net, DC of 0.8026 and JI of 0.6828, and SegNet, DC of 0.8052 and JI of 0.6899. These developments highlight the strength of attention mechanisms in deep learning-based eye segmentation, making the proposed method a suitable choice for applications requiring high-precision ocular image analysis. The high accuracy obtained with the proposed model allows for more precise analysis of eye diseases in medical diagnosis processes. Therefore, more effective decision support systems can be developed in both security and health areas when considered for the field of biometric verification. Keywords - Eye segmentation, deep learning, attention mechanisms, convolutional neural networks, biometric recognition.