Reliable person identification using a novel multibiometric image sensor fusion architecture

Abstract Biometric and multimodal biometric sectors have made major advances in recent years. Usually, forensics, security, and privacy are the three areas where this is exhibited. It is frequently impossible to reach a greater identification rate with unimodal biometric systems, even the best ones. Unimodal biometric systems have several drawbacks, including no universality, a higher rate of false acceptance, and lower rates of true acceptance. Multimodal biometric systems get around these drawbacks. As there are many pieces of evidence for the same identification, a more reliable recognition performance can be attained. The multimodal biometric system that uses iris and fingerprints is the main topic of the work discussed in this paper. The Gabor Filters (GF) are used to extract the unique textual characteristics of the iris and fingerprint. These unimodal characteristics are combined utilizing the Mahalanobis distance technique using a new feature-level fusion method. The system is trained to utilize the retrieved feature using a hybrid random forest classifier with a channel-wise convolutional neural network (HRFC-CWCNN) based learning technique. Using the CASIA iris dataset and genuine fingerprint dataset, the performance of the suggested methods is verified and evaluated against that of other algorithms. It is clear from the simulation outcomes which our method outperforms previous techniques in terms of recognition rate and false rejection rate.