Image Deblurring and Denoising using Color Priors


Neel Joshi C. Lawrence Zitnick Richard Szeliski David Kriegman
Microsoft Research
Microsoft Research Microsoft Research UCSD

To appear at
CVPR 2009


(a) Many sharp edges that can blur to match an observed blurred (and potentially noisy) edge (in tan). The sparse prior always prefers the smallest intensity gradient that is consistent with the observation (in red). Our method picks the edge that is more likely given the dominant primary and secondary colors in the pixel’s neighborhood. (b) The thin blue areas between the white letters are deconvolved to a mid-level blue with the sparse prior. With our method the blue spaces are more distinct and edges are sharper.




Image blur and noise are difficult to avoid in many situations and can often ruin a photograph. We present a novel image deconvolution algorithm that deblurs and denoises an image given a known shift-invariant blur kernel. Our algorithm uses local color statistics derived from the image itself as a constraint in a unified framework that can be used for deblurring, denoising, and upsampling. A pixel’s color is required to be a linear combination of the two most prevalent colors within a neighborhood of the pixel. This two-color prior has two major benefits: it is tuned to the content of the particular image and it serves to decouple edge sharpness from edge strength. Our unified algorithm for deblurring and denoising out-performs previous methods that are specialized for these individual applications. We demonstrate this with both qualitative results and extensive quantitative comparisons that show that we can outperform previous methods by approximately 1 to 3 DB.


Paper (Preprint)

Adobe Acrobat PDF (6.41 MB)

Supplementary Results



Copyright 2009 by Neel Joshi, UCSD, and Microsoft Research