Deblurring Network for Accurate Facial Landmark Detection

We present a face image deblurring framework for accurate facial landmark detection. In face recognition application, performance is strongly affected by the detection accuracy of facial landmarks. Therefore, when developing a facial image deblurring algorithm, it is important not only to improve the restored image quality, but also to improve the landmark detection accuracy of structural parts of the face. In this work, we propose a simple but efficient deblurring network using Wavelet loss and Landmark-error related losses. Experimental results show that the proposed network has superior objective/subjective performance compared to the existing methods.