how to handle occlusion for face recognition?
If there some occlusion on face (such as sunglasses, mask, scarf), the recognition rate will decrease steeply? How to handle this case?
I have do a survey, the PCA reconstruction method have been mentioned by some literature. I don't know whether or not this method is the state-of-the-art method. Is there other method for address occlusion problem?
Part based modeling could do the trick OR simply retrain a face model using a dataset that contains these smaller occlusions. Keep in mind that the supplied model has training data where people just look straight into the camera without a single obstruction. You could add however a glasses detector and a scarf detector to increase the chance of a face being there.