A two step method to recover occluded part of face

In this paper, we propose an approach to recover the occluded part of a face image with the help of prior knowledge learned from a large face database. Firstly, a global model is applied to represent the relation-ship between the occluded part and non-occluded part of a face image. Then, a local compensation method is used to estimate high frequency content. A high quality face image is achieved by integrate both global and high frequency contents. Experiments show that our approach can recover a naturally face image with well preserved global and local detail information.