Occlusion-Robust Face Recognition Using Iterative Stacked Denoising Autoencoder

摘要

This paper investigates how to recognize faces with partial occlusions using iterative stacked denoising autoencoder (ISDAE). We introduce a mapping-autoencoder (MAE) for occlusion detection, which requires no prior knowledge of occlusion. Inspired by stacked denoising autoencoder (SDAE)'s capability to learn patterns from noisy data, we propose a novel iterative structure of SDAE for occluded faces restoration. Deep neural network (DNN) is used for final recognition. Compared with the state-of-the-art approaches (e.g. sparse representation), ISDAE achieves competitive results under serious occlusion conditions.