Occlusion?invariant face recognition using simultaneous segmentation

When using convolutional neural network (CNN) models to extract features of an occluded face, the part will inevitably be embedded into representation just as with other facial regions. Existing methods deal face recognition either by augmenting training dataset synthesized faces or segmenting occlusions first and subsequently recognize based on unoccluded Instead, simultaneous occlusion segmentation is developed make most these correlated two tasks. This inspired phenomenon that corrupted are traceable within a CNN trained segment parts in images. Specifically, invariant deep (SOIDN) proposed contains simultaneously operating networks coupled mask adaptor module their bridge learn features. The SOIDN jointly supervised classification losses aiming obtain (1) features, (2) segmentation, (3) feature weighs reliability Experiments (e.g. LFW-occ) real AR) demonstrate outperforms state art for verification identification.