Joint Face Image Restoration and Frontalization for Recognition

In real-world scenarios, many factors may harm face recognition performance, e.g., large pose, bad illumination,low resolution, blur and noise. To address these challenges, previous efforts usually first restore the low-quality faces to high-quality ones then perform recognition. However, most of methods are stage-wise, which is sub-optimal deviates from reality. this paper, we all challenges jointly for unconstrained We propose an Multi-Degradation Face Restoration (MDFR) model frontalized given under arbitrary facial poses, with three distinct novelties. First, MDFR a well-designed encoder-decoder architecture extracts feature representation input image restores it counterpart. Second, introduces pose residual learning strategy along 3D-based Pose Normalization Module (PNM), can perceive gap between initial its real-frontal guide frontalization. Finally, generate images by single unified network, showing strong capability preserving identity. Qualitative quantitative experiments on both controlled in-the-wild benchmarks demonstrate superiority over state-of-the-art frontalization restoration.