PROJECT TITLE :

Accelerating GMM-Based Patch Priors for Image Restoration Three Ingredients for a 100_ Speed-Up

ABSTRACT:

The goal of picture restoration is to restore a clear image from a smudged one. In order to restore natural visuals, the EPLL algorithm first applies a Gaussian mixture model (GMM) to the patches in the original image. However, EPLL's high runtime complexity means it is unsuitable for most practical applications, even if it is quite effective for image restoration. It is proposed in this study that three approximations to the original EPLL method can be developed. Our fast-EPLL (FEPLL) technique achieves two orders of magnitude faster performance than EPLL while sacrificing almost no image quality in the process (less than 0.5 dB). An array of inverse issues, including denoising, deblurring, super-resolution, inpainting, and devignetting, show the usefulness and adaptability of our algorithm. For all of the degradations listed above, FEPLL is the first approach that can be used without any extra code optimizations, such as CPU parallelization or GPU implementation, to restore a 512x512 pixel image in less than 0.5s.