LTT-GAN: Looking Through Turbulence by Inverting GANs

In many applications of long-range imaging, we are faced with a scenario where person appearing in the captured imagery is often degraded by atmospheric turbulence. However, restoring such images for face verification difficult since degradation causes to be geometrically distorted and blurry. To mitigate turbulence effect, this paper, propose first mitigation method that makes use visual priors encapsulated well-trained GAN. Based on priors, learn preserve identity restored spatial periodic contextual distance. Such distance can keep realism from GAN while considering difference at network learning. addition, hierarchical pseudo connections proposed facilitating identity-preserving learning introducing more appearance variance without changing. Extensive experiments show our significantly outperforms prior art both quality accuracy results.