A Segmentation Scheme For Robust Iris Based On Improved U-Net
B. Pavan Kumar, V. G. Bhavani, N. Ch. S. Prasad, G. P. Kumar, G. Roop Kumar, K. Shanmukh · Advances in computer science research · 2024
The reliance of the iris recognition system on high-quality iris segmentation creates a strong foundation for future iris recognition research and greatly improves the efficiency of iris identification.By using the same datasets for training and testing, we were able to obtain the top network model, FD-UNet.This network architecture combines elements from U-Net and four others that have proven effective.By switching from original convolution to dilated convolution, the FD-UNet improves picture processing by extracting more global features.Datasets such as UBIRIS.v2for visible light illumination, CASIAiris-interval-v4.0 and ND-IRIS-0405 for near-infrared illumination, and others were utilised to evaluate the proposed method.Our model hit f1 scores of 97.36%, 96.74%, and 94.81% on the CASIA-iris-interval-v4.0, ND-IRIS-0405, and UBIRIS.v2datasets, respectively.Our network model outperforms the competition with a reduced error rate, according to the trial data.It is quite reliable and performs admirably with iris datasets that use both visible light and near-infrared lighting.