DEEP LEARNING-BASED FACE-IRIS RECOGNITION WITH CAPSULE NEURAL NETWORK (CAPSNET) FOR BIOMETRIC ATTENDANCE SYSTEM

Abstract

A model of biometric-based attendance system will be presented in this paper, which will utilize face recognition and iris recognition in order to make the attendance system more accurate and achieve greater level of security in the verification of the subject identity. Following the introduction of the proposed framework, early on, the acquisition of the CASIA-Iris-Distance Dataset is conducted, and subsequently, the basic pre-processing methods such as Hough Transform of Circles to segment the image, Min-Max normalization to scale the features, and Canny edge detection to visualize the iris pattern well is implemented. During the extraction of facial features where the size has to be reduced and the deep salient facial features have to be extracted, a Sparse Autoencoder (SAE) is used. These features extracted by the face and the iris modalities are then merged and placed into a Capsule Network, i.e., the Hyper Capsule Neural Network (HCNN), which maintains spatial hierarchies and shows a robust recognition performance regardless of small data changes. The various performance metrics are employed in determining that this system is workable and dependable in the real world. As a result of the combination of the recognition outputs, the system automatically predicts and records individual attendance. The proposed hybrid model (SAE+HCNN) through comparative analysis of performance is more precise and the accuracy of this proposed model is approximately 95% as compared to that of the SAE+CNN in which the accuracy is approximately 93-94%. The results provided above indicate that combination algorithm is very practical to reduce the risk of spoofing and boost the recognition accurateness in comparison to common single modal or multimodal early fusion systems.