RAG‐Net: ResNet‐50 attention gate network for accurate iris segmentation

Abstract Iris segmentation is an important step in the process of iris recognition. Iris images collected under non‐cooperative conditions always contain various noise, which is a challenge for iris segmentation. Most U‐Net‐based methods have made great achievements in iris segmentation. However, this architecture lacks of focusing on target structures of varying shapes, and robustness in segmenting objects with significant shape variations. In this paper, we propose RAG‐Net: an efficient iris segmentation method based on deep learning. In contrast to many previous convolutional neural network (CNN)‐based iris segmentation methods, we adopted the attention gate (AG) mechanism and ResNet‐50 in the U‐Net architecture to improve iris segmentation accuracy, the AG module was included in the skip connection part of the RAG‐Net architecture to further identify salient feature regions and prune feature responses, which preserve only the activations relevant to the required information, and the ResNet‐50 module was used to improve the robustness of the segmentation performance. Using this model, efficient iris segmentation in a non‐cooperative environment can be realized. The proposed method was trained and evaluated using the CASIA.v4‐distance, CASIA.v4‐thousand, UBIRIS.v2, and MICHE‐I databases. From the view of the segmentation results, the proposed RAG‐Net is one of effective architecture in iris segmentation methods.

Abstract licence: CC BY-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.