Real-time masked face recognition using deep learning-based double generator network

The COVID-19 outbreak has spread rapidly worldwide since 2019. This pandemic has complicated and intricate human existence, and thousands have died from it. Due to the spread of coronavirus, people wear masks while going outside. Consequently, the system cannot identify their faces while wearing the masks. This issue can be overcome by introducing a system that recognizes masked faces of random people trained with 100 images taken from the Internet. This paper presents a novel deep learning-based double generator network to precisely identify the face behind the mask images. Initially, the gathered images are split into low- and high-frequency components using 2D-stationary wavelet transform (2D-SWT). Afterward, the Haar cascade classifier was implemented to capture the masked image biometrics to recognize the individual faces. The proposed double generator network involves two modules: edge generation and image reconstruction. The first modules consist of dilated convolutional for retrieving the relevant features from the masked face images created on the generated edges. The generated edges are reconstructed using the reflection of generated edges in the second module. Finally, the output images are reconstructed to identify the masked face. From the simulation results, the proposed framework showed effective performance based on the network parameters. The proposed network attains an accuracy of 97.08% for masked face recognition which demonstrates it achieves higher accuracy than the prior frameworks.