Correlation Filter of Deep Features for Robust Face Recognition

In this paper, a robust face recognition method combined correlation filter with deep learning is proposed. The deep features is first extracted for all the face images of one subject, and then merged to a correlation vector through Minimum Average Correlation Energy (MACE) filter. During recognition, the probe image only needs to be compared with the correlation filter rather than the raw image features. The proposed algorithm combines the advantages of both deep learning and correlation filter. Experiments show that the recognition result of our method is more accurate and the computation time is shorter compared with other state-of-the-art algorithms.