Iris Segmentation Using Face Images For Authentication Using Hybrid Approaches Of Fusion Of SVM and Deep Learning

Abstract

Biometric authentication systems have emerged as a cornerstone of secure identity verification due to their reliability and robustness. Among these, iris recognition stands out for its accuracy and resistance to forgery. However, integrating iris segmentation from face images poses significant challenges, such as occlusion, variations in illumination, and diverse facial orientations. This paper introduces a novel hybrid approach that combines Support Vector Machines (SVM) and deep learning techniques to address these issues and enhance segmentation accuracy. The proposed framework starts with pre-processing of facial images to extract the region of interest ROI, namely in the iris region. Coarse segmentation and key feature extraction is performed with the help of a Convolutional Neural Network (CNN). Subsequently, SVM refines these outputs by optimizing pixel-level boundary detection, leveraging its strength in classification tasks. The fusion of these two techniques is implemented at the decision level, integrating the outputs of CNN and SVM to produce a highly precise segmented iris region. This dual-layered methodology ensures robust segmentation, even under challenging conditions. Standard sets, such as CASIA and UBIRIS are used to test the framework with improved performance compared to traditional methods[1]. Accuracy, IoU, and F1-score are shown to demonstrate the effectiveness of the hybrid strategy. In addition, the adaptability of the system with different datasets proves that it can be well utilized in practice. This study moves biometrics identification a step forward with efficient methods and scalable approaches of iris segmentation through hybrid techniques. The synergy between SVM and deep learning exploits the benefits of both methods, and forms a foundation of further research into safe authentication systems. The proposed framework can be extended to other biometric modalities, offering a versatile tool for enhancing authentication accuracy and reliability.