Facial occlusion, such as sunglasses, scarf, mask etc., is one critical factor that affects the performance of face recognition. Unfortunately, faces with occlusion are quite common in the real world, especially in uncooperative scenario. In recent years, regression analysis becomes a hotspot of dealing with face recognition under different illuminations and facial occlusions. The basic idea of regression analysis is to recover clean images from degraded images or occluded images by using the clean training samples. Then the reconstructed images are used for face recognition. However noise would be introduced in the recovery procedure. So whether reconstructed image help face recognition is still worth studying. Note that the residual image which is a difference between the raw and reconstructed image containing most of the occluded information. We can use it for occlusion detection. In this paper we make two contributions: i) we present a new occlusion detection method by combining the information of both raw image and residual image; ii) we empirically show that using the non-occluded part for face recognition has a better result than using reconstructed image.

No abstract

No abstract

No abstract