Ask more Info

Blind Image Deblurring Via Residual Blur Elimination

Images may be blurred due to various reasons. Defocus-blur may be caused by incorrect camera settings, poor lens selection or scenes with depth variations exceeding the depth of field of the camera. Motion-blur can appear due to motion in the scene, camera movements and vibrations.
Image deblurring, also known as `image restoration’ and `image deconvolution’ , is a digital image processing procedure, intended to recover the sharp latent image from a given blurred image. If the mechanism that caused the blur is fully known at the deblurring stage, and can be characterized by a given blur kernel or a point spread function (PSF), the deblurring problem is referred to as non-blind. Otherwise, when the specifics of the blur process are unknown, the problem is called blind deblurring.
Image deblurring is notoriously difficult, characterized as a mathematically ill-posed inverse problem – meaning that the given blurred image can be explained by numerous different latent images, without an obvious way to choose the correct one.  In the simpler non-blind case, classical algorithms, such as the inverse, pseudo-inverse and Wiener filters, are inadequate. Significant progress in non-blind deblurring has been made in the last 10-15 years, mostly based on mathematically-sophisticated variational approaches.
The blind formulation of the deblurring problem is still much harder, as it is doubly ill-posed – with uncertainty relating both to the latent image and the blur kernel. Solid successes have appeared only in the last 5 years or so, but the few available methods are computationally demanding and require careful, case by case tuning.
We achieved a breakthrough in blind image deblurring, allowing reliable restoration with minimal tuning and reasonable computational requirements. Potential applications include the correction of both defocus and motion blur, in diverse domains such as smartphone photography and drone-based aerial imaging.


“Blind” image restoration examples
The suggested approach is readily applicable to color and IR images.