LAUSR

Iris biometrics is one of the frequently used biometrics for security purposes, and it provides the appropriate tools for identifying humans. Deep learners when used along with iris recognition can improve accuracy, automatic learning, and generalization ability. Despite the advantages of deep learners, some problems like time complexity and computational efforts exist in deep learners. In order to solve these obstacles, this paper intends to develop the intelligent iris recognition model based on the concepts merged with multi-objective feature selection and deep learning. The main phases of the proposed model involve (a) Pre-processing, (b) Iris Segmentation, (c) optimal ORB point selection and feature vector generation, and (d) optimal recognition. Here, the pre-processing of the image is performed by filtering and contrast enhancement techniques. Further, the iris segmentation is done by few approaches such as reflection moving, iris localization, and Hough-transform-based segmentation. Once the segmentation of the iris is finished, feature extraction is carried out by optimized Oriented FAST and rotated BRIEF (ORB) features concerning on multi-objective function. Finally, the optimized Convolutional Neural Network (CNN) is used for iris recognition. Here, the hybrid algorithm with the integration of two meta-heuristic algorithms like Elephant Herding Optimization (EHO), and Whale Optimization Algorithm (WOA) called Elephant Herding with Whale Optimization Algorithm (EH-WOA) is utilized for performing the optimized ORB feature selection and optimized CNN. The experiments are conducted on a benchmark datasets like IIT Delhi (IITD) Iris database, and MMU iris dataset to analyze the performance of the proposed model over the different network structures for accurate iris recognition. Through the experimental analysis, the proposed optimized CNN + EH-WOA had better values in terms of accuracy, which is 0.98% higher than SVM, 0.92% higher than KNN, 0.88% higher NN, and 0.85% higher than CNN respectively for the learning % as 75 for MMU iris dataset. Similarly, several performance measures have considered for showing the efficiency of the designed model.