Masked face recognition via dual-branch convolutional self-attention network

Weiguo Wan, Runlin Wen, Linghan Deng, Yong Yang

Abstract: Highlights•A dual-branch network is proposed based on the convolutional self-attention architecture to improve the MFR accuracy.•An MCSAM is constructed to extract local and global features for MFR, and an MMHSA is designed to enhance global features.•An UFAM is designed to improve MFR performance by assigning higher weights to the upper face features.