CNN-Based Restoration of a Single Face Image Degraded by Atmospheric Turbulence

Atmospheric turbulence significantly affects imaging systems which use light that has propagated through long atmospheric paths. Images captured under such condition suffer from a combination of geometric deformation and blur. We present a deep learning-based solution to the problem of restoring a single turbulence-degraded face image where the amount of geometric distortion and blur at each pixel location is first estimated in terms of variance maps using two separate networks. The estimated variance maps are then used by the Turbulence Distortion Removal Network (TDRN) to restore the image. Furthermore, a confidence-guided image gradient-based loss is proposed to train TDRN. Comprehensive experiments on synthetic and real face images show that the proposed framework is capable of alleviating blur and geometric distortion caused by atmospheric turbulence, and can significantly improve the visual quality. In addition, an ablation study is performed to demonstrate the improvements obtained by different modules in the proposed method.