A paper to appear at the IEEE (Institute of Electrical and Electronics Engineers) International Conference on Computer Vision Workshops 2017 introduced a framework for performing facial recognition of partially covered or disguised faces, such as those of protesters who wear masks and scarves to hide their identity.

The deep learning framework is designed to detect 14 facial key points which are then used for facial recognition. The authors are Amarjot Singh, Department of Engineering, University of Cambridge, UK;
Devendra Patil, National Institute of Technology, Warangal, India; G Meghana Reddy, National Institute of Technology, Warangal, India; and SN Omkar, Indian Institute of Science, Bangalore, India.

According to the paper, the framework, which employs Spatial Fusion Convolutional Network, outperforms state-of-the-art methods on key-point detection and face disguise classification. 

To see the paper, visit arxiv.org/pdf/1708.09317v1.pdf