Enhancing Face Recognition Accuracy: A Novel Approach Using EfficientNet B7 and Triplet Loss Learning

Face recognition has long been used as an authentic and reliable source for a wide range of authentication systems. But, to maintain the accuracy of this technology, the facial landmarks must be clearly captured, both for training and testing purposes in the recognition models. It becomes especially challenging for the system to deliver performance when parts of faces are covered, for example in cases where a person puts on a face mask. In this work, a robust face recognition model is utilized and have delved into the changes to deliver decent results even when parts of the face are occluded. The proposed work employed the B7 version of the EfficientNet model and combined it with online learning, triplet loss learning, to get an accuracy of 95.85% on a dataset containing around 1800 test images. Simulation results have been depicted to show the efficacy of the proposed method over the previously implemented models to benchmark their performances.