Deblurring atmospheric turbulence degraded images using an isolate edges prior

Atmospheric turbulence affects the imaging system at a long distance, which causes time-varying blur. Although the blur kernel is unknown, we propose an algorithm to estimate Optical Transfer Function (OTF) for long-exposure atmospheric turbulence blurred images. In this paper we present a novel image prior-isolate edges prior to predict a sharp 'vision' of degraded image edges, and utilize the two images to solve for the OTF. In this prior, the isolate edges of the image gradients are modeled by parametric exponential functions. We transform the estimated OTF to blur kernel and modify it using a maximum-a-posteriori model. Finally, an effective non-blind deconvolution is employed to obtain the output image with the modified blur kernel. The kernel we get is anisotropic. Numerical experiments suggest that this algorithm can obtain accurate blur kernel from real images and is able to alleviate blur, recovering details, sharpening edges of the scene and improving visual quality significantly.