Stochastic Blind Motion Deblurring

Abstract

Blind motion deblurring from a single image is a highly under-constrained problem with many degenerate solutions. A good approximation of the intrinsic image can therefore only be obtained with the help of prior information in the form of (often non-convex) regularization terms for both the intrinsic image and the kernel. While the best choice of image priors is still a topic of ongoing investigation, this research is made more complicated by the fact that historically each new prior requires the development of a custom optimization method. In this paper, we develop a stochastic optimization method for blind deconvolution. Since this stochastic solver does not require the explicit computation of the gradient of the objective function and uses only efficient local evaluation of the objective, new priors can be implemented and tested very quickly. We demonstrate that this framework, in combination with different image priors produces results with PSNR values that match or exceed the results obtained by much more complex state-of-the-art blind motion deblurring algorithms.

Processing and analysis of motion-blur images in light scattering flow cytometry

“…Approximating the blur kernel in the motion-blur images is the fundamental challenge for deblurring images. We apply natural image priors and Bayesian approach to obtain the blur kernel in the motion-blur images, then reconstruct the images by a standard deconvolution algorithm 17,18 .…”

Blind motion deblurring using improved DeblurGAN

“…Xiao et al. [12] proposed a blind inverse convolution stochastic optimization method, which does not explicitly calculate the gradient of the objective function and only uses an effective local evaluation of the objective, and can therefore be implemented and tested quickly. Lazareva et al.…”

“…Tao et al [11] proposed a scale-recurrent network (SRN) to recover the sharp edges of blurred images. Xiao et al [12] proposed a blind inverse convolution stochastic optimization method, which does not explicitly calculate the gradient of the objective function and only uses an effective local evaluation of the objective, and can therefore be implemented and tested quickly. Lazareva et al [13] used a random forest to learn retinal images, using Zernike coefficients to represent a mapping of the blur kernel space, to reconstruct adaptive optical high-resolution retinal images [14].…”