Multilevel Authentication System Based on Periocular Features Using Deep Learning Algorithm

Abstract- The iris recognition biometric technique faces limitations primarily due to the high costs associated with optical equipment and the inconvenience experienced by users. As an alternative, periocular-based methods offer a viable solution for biometric authentication, as they do not necessitate costly devices. Furthermore, the data obtained from these methods are valuable for biometrics since they capture features such as eyelashes, eyebrows, and eyelids. However, traditional periocular-based biometric authentication techniques rely on restricted sets of features based on the chosen feature extraction method, leading to comparatively subpar results. Consequently, we introduce a deep-learning approach that makes full use of the diverse features present in periocular images. This method preserves the mid-level features from the convolutional layers and selectively incorporates those that are most beneficial for classification. We evaluated the proposed approach against prior methods using both publicly available and self-gathered datasets. The results of the experiments indicate an equal error rate of less than 1%, outperforming earlier techniques. Additionally, we present a novel methodology to assess whether mid-stage features have been effectively utilized. As a result, it was demonstrated that this strategy, which leverages mid-level features, significantly enhances the performance of feature extraction within the network.
DOI: 10.61137/ijsret.vol.9.issue1.132