Improving Masked Face Recognition Using Dense Residual Unit Aided with Quadruplet Loss

All over the world, people are wearing face masks and practising social distancing to protect themselves against the Coronavirus disease (COVID-19). The need for contactless biometric systems has increased to avoid the common point of contact. Among contactless biometric systems, facial recognition systems are the most economical and effective ones. Conventional face recognition systems rely heavily upon the facial features of the eyes, nose, and mouth. But due to people wearing face masks, the important facial features of the nose and mouth get hidden under the mask, resulting in degraded performance by the facial recognition systems on masked faces. In this paper, we propose a Dense Residual Unit (DRU) aided with Quadruplet loss on top of existing facial recognition systems. This solution tries to unveil the masked faces by producing embeddings for masked faces, which are similar to embeddings of unmasked faces of the same identity but different from embeddings of different identities. We have evaluated our method using two pre-trained facial recognition models' backbones, i.e. Resnet-101 and MobileFaceNet, and upon two datasets, among them, one is a real-world dataset, i.e. MFR2, and one is a simulated masked face dataset i.e. masked version of LFW. We have achieved improvement in the performance of masked face recognition in terms of False Match Rate, False Non-Match Rate, Fisher Discriminant ratio, and Equal Error Rate. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Recognition of Faces Wearing Masks Using Skip Connection Based Dense Units Augmented With Self Restrained Triplet Loss

Facial recognition-based systems are the most efficient and cost-effective of all the contactless biometric verification systems available. But, in the COVID-19 scenario, the performance of available facial recognition systems has been affected badly due to the presence of masks on people's faces. Various studies have reported the degradation of the performance of facial recognition systems due to masks. Therefore, there is a need for improvement in the performance of currently available facial recognition algorithms. In this research, we propose using Skip Connection based Dense Unit (SCDU) trained with Self Restrained Triplet Loss, to handle the embeddings produced by existing facial recognition algorithms for masked images. The SCDU is trained to make facial embeddings for unmasked and masked images of the same identity similar, as well as, embeddings for unmasked and masked images of different identities dissimilar. We have evaluated our results on the LFW dataset with synthetic masks as well as the real-world masked face recognition dataset, i.e., MFR2 and achieved improvement in verification performance in terms of Equal Error Rate, False Match Rate, False Non-Match Rate, and Fisher discriminant ratio. © 2022 IEEE.

Attention Based Residual Network for Effective Detection of COVID-19 and Viral Pneumonia

The current coronavirus (COVID-19) pandemic has led us to the healthcare, global poverty and socioeconomic crisis. One of the most significant task in this pandemic is to accurately and efficiently diagnose the COVID-19 patients and to monitor them to make prompt decisions and take appropriate actions for their monitoring, management and treatment. The early diagnosis of COVID-19 was a very troublesome and difficult challenge that CAD (Computer-Aided Diagnosis) methods successfully tackled. The CXR (chest X-ray) method proved to be a very low-cost and effective alternative to Computed Tomography (CT) scan and Real Time Polymerase Chain Reaction (RT-PCR) test, which were previously the most commonly used methods for COVID-19 diagnosis. Till now, very few CAD based techniques have been proposed to effectively detect COVID-19, but their efficiency is limited due to a number of factors. In this study, we have proposed a deep learning model using the Convolutional Block Attention Module with ResNet32. For training the model, Kaggle's dataset containing CXR images has been used. The dataset contains a total of 3886 images. Moreover, 64% of data has been used for training, 20% for testing and 16% for validation. We have experimented with different CNN architectures with different approaches like Transfer Learning, Data Augmentation and attention module. With 97.69% accuracy, the ResNet32 with attention module outperformed other architectures and approaches, improving the baseline network efficiency. This promising and efficient classification accomplishment of our proposed methodology demonstrates that it is well suited for CXR image classification in COVID-19 diagnosis in terms of both accuracy and cost. © 2021 IEEE.