Edge and identity preserving network for face super-resolution

Face super-resolution (SR) has become an indispensable function in security solutions such as video surveillance and identification system, but the distortion facial components is a great challenge it. Most state-of-the-art methods have utilized priors with deep neural networks. These require extra labels, longer training time, larger computation memory. In this paper, we propose novel Edge Identity Preserving Network for SR Network, named EIPNet, to minimize by utilizing lightweight edge block identity information. We present extract perceptual information, concatenate it original feature maps multiple scales. This structure progressively provides information reconstruction aggregate local global structural Moreover, define loss preserve of images. The compares distributions between images their ground truth recover identities addition, provide luminance-chrominance error (LCE) separately infer brightness color LCE method not only reduces dependency dividing also enables our network reflect differences two spaces RGB YUV. proposed facilitates elaborately restore generate high quality 8x scaled structure. Furthermore, able reconstruct 128x128 image 215 fps on GTX 1080Ti GPU. Extensive experiments demonstrate that qualitatively quantitatively outperforms challenging datasets: CelebA VGGFace2.