Blind restoration of space-variant Gaussian-like blurred images using regional PSFs

In this paper, we propose an alternative de-blurring method to remove Gaussian blurs (that approximate distortions such as out-of-focus and long atmospheric path). Our research focuses on images that contain space-variant blurs, as well as focused objects within the same image. We propose to use estimated depth map-based segmentation for space-variant blind image de-blurring. Unlike previous studies that assume PSF or use one calculated PSF for the entire image, our method composes re-blurred images using different calculated PSF’s and combining the restored images to one restored image. The proposed method contains several parts: blur map estimation based on edge-width analysis, division of the image into several layers according to the blur severity, and extraction of the point-spread function (PSF) that characterizes the blur at each layer. Next, each layer is de-blurred using its specific PSF and the total variation method; the de-blurred layers are combined forming the final restored image. A main challenge is to prevent distortions of the sharper objects in the image while simultaneously sharpening the blurrier objects.