Human Iris Detection Under Multiple Occlusion Using Makesense AI and Yolo.V5

The uniqueness of the iris without any deterioration as a result of ageing makes it preferable to other detection systems. However, the robustness of iris detection systems is mostly questioned due to noise such as eyelashes, eyelids, illumination variation, and blurred edges. The need to improve the localization of the iris region keeps growing every day. Some recent studies have proposed conformal geometric algebra (CGA) and the region-based convolutional neural network (R-CNN) to address the segmentation issues on noisy iris images. The CGA still has a problem resolving iris images that contain eyelashes and eyelids. The R-CNN had issues resolving noise in high-quality images with clear iris boundaries. To improve on the issue resulting from existing works, this study proposed a YOLO V5 model for detecting iris on noisy iris images. First, Makesense AI, an image segmentation tool was used to localize the iris region. Then the YOLO V5 model was used to extract iris features using the CBS and subsequently detect the iris. Experiments were conducted with IITD, CASIA V1, and MMU iris datasets. The proposed model obtained an accuracy of 100% and a mean average precision (mAP) of 99.5% on IITD datasets, an accuracy of 100% and mAP of 99.4% on CASIA iris images and an accuracy of 100% and mAP of 99.6% on MMU iris images.