NIPS2009_0341_slide - min x i =1 2 x ⊕ k − y i j =1 | x ⊕ f j i | • Existing methods slow ILRS takes ~1hr for 1 megapixel image Our

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
Fast Image Deconvolution using Hyper Laplacian Priors Dilip Krishnan, Rob Fergus Courant Institute, New York University Hyper-laplacian priors are good models of gradient distributions in natural images - Used in deblurring, denoising, super-resolution etc. Consider the deconvolution problem with sparse gradient prior ( < 1). Want , given blurry and kernel N λ 2 J α x k α y α
Background image of page 1
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: min x i =1 ( 2 ( x ⊕ k − y ) i + j =1 | ( x ⊕ f j ) i | ) • Existing methods slow: ILRS takes ~1hr for 1 megapixel image Our approach: • Splitting mechanism gives 2 sub-problems: (i) quadratic in b solve with FFT (ii) 1-D problem in auxiliary b Solve with look-up table • Achieve 2-3 orders of magnitude speed-up over IRLS x x x w Likelihood Prior...
View Full Document

This note was uploaded on 02/12/2010 for the course COMPUTER S 10586 taught by Professor Jilinwang during the Fall '09 term at Zhejiang University.

Ask a homework question - tutors are online