Masked Face Detection Based on Locally Nonlinear Feature Fusion

Realizing that features with strong discrimination are both needed for the generation and discrimination of candidate regions in the masked face detection, LNFF-Net (Locally Nonlinear Feature Fusion-based Network) is proposed. To highlight the feature from the face region and suppress the background region, this method nonlinearly fuses the visual saliency map and the heat map, which is extracted from a light fully convolutional network (FCN). On the other hand, through transferring the Fast R-CNN based multi-objective detection to single masked face detection, the structure of candidate region discrimination using convolutional network is optimized. Experimental results show that the proposed algorithm has better detection accuracy than other method.