Semantic Segmentation of Iris from Pterygium for Biometric Process using U-Net
Ambhara Putri Rie Kikhaawa, Khoerun Nisa Syaja’ah, Yudha Satya Perkasa, Rin Rin Nurmalasari, Weihua Yang · 2024
There is an alternative advanced identification system called biometrics. Biometrics uses unique human body parts to identify a person. One part of the human body considered unique and can be utilized by a biometric system is the iris. The use of biometrics on the eye's iris is also known as the Iris Recognition System (IRS). Segmentation is one of the stages of the IRS that is very influential in the accuracy of individual identification. The segmentation stage also has challenges, such as pterygium tissue occlusion in pterygium sufferers. In this research, segmentation was carried out for these challenges using the U-Net method. U-Net is a method that can be used for biomedical image segmentation tasks because it only requires a small amount of label data and a very reasonable time for model training. Segmentation of pterygium sufferers shows subjectively inaccurate prediction image results at any epoch. However, the evaluation metrics show increasingly better results as the epoch increases. The best results were obtained by epoch 300, namely an IoU Coeff of 0.96 and a Dice Coeff of 0.92.