A Convolutional Network-based Applied Approach to Frontal Facial Recognition and Solutions for Occluded Images

Facial recognition technology is constantly being used in the most diverse sectors, from facial expression analysis to measure customer satisfaction in stores to a police instrument for identifying people. To summarize, from an image obtained by a camera, the technique identifies faces contained in that photo and compares them with a database of faces previously registered in the system. Based on advances in terms of algorithms and processing of hardware obtained in recent years, it was possible to provide solutions for verification and facial recognition through modern cell phones. The present work addresses the problem of facial recognition analysis that has the potential to scale for applications based on server processing. Using Histogram of Oriented Gradients to extract face features and the FaceNet Convolutional Neural Network this study brings results in different public (Labeled Face in the Wild and CelebFaces) and private databases. This study obtained satisfactory results with an accuracy of 90% in the best cases for private databases. As part of the work, this study also sought to evaluate the effect of partial occlusion on faces from the use of face masks because of the Covid-19 pandemic scenario, obtaining satisfactory results above 80%.