A Novel Iris Recognition System Development Using Convolutional Neural Network

Biometric devices have made tremendous advancements in person identification and authentication, contributing considerably to personal, governmental, and international security. In full iris recognition systems, iris segmentation algorithms play a significant role and have a direct impact on the outcomes of iris revivification and recognition. However, when used on noisy iris datasets recorded in unrestricted environments, standard iris segmentation algorithms exhibit low flexibility and are insufficiently resilient. Furthermore, there is presently no significant segmentation based on iris datasets methods cannot fully use the potential of convolutional neural networks (CNNs). The key contribution of the following paper: Initially, we suggest a CNN-based architecture. A dense-fully convolution network is thus named (DFCN) for iris segmentation and uses well-known optimizer techniques such as dropout and batch normalization (BN). Second, using the LabelMe software programmed, we label the areas that cover the iris regions because they are absent from the IITD iris and CASIA-Interval-v4 public ground-truth masks datasets. The encouraging outcomes of trials based on the IITD, UBIRIS, and CASIA-Interval-v4 are the last. In terms of a variety of metrics, including accuracy, precision, recall, f1 score, nice1 and nice2 error scores, the iris segmentation network proposed in this paper outperforms all conventional and the majority CNN-based iris segmentation algorithms, demonstrating the robustness of our suggested network. This is shown by databases of the V2 iris taken under various circumstances. The findings clearly suggest that convolution-based methods may be used to create that spoofing detection system resistant to known assaults and perhaps easily adapt to future image-based attacks.