Image restoration by matching gradient distributions

IEEE Transactions on Pattern Analysis and Machine Intelligence

Taeg Sang Cho1 C. Lawrence Zitnick2 Neel Joshi2 Sing Bing Kang2 Richard Szeliski2 William T. Freeman1

1 Massachusetts Institute of Technology

2 Microsoft Research


The restoration of a blurry or noisy image is commonly performed with a MAP estimator, which maximizes a posterior probability to reconstruct a clean image from a degraded image. A MAP estimator, when used with a sparse gradient image prior, reconstructs piecewise smooth images and typically removes textures that are important for visual realism. We present an alternative deconvolution method called iterative distribution reweighting (IDR) which imposes a global constraint on gradients so that a reconstructed image should have a gradient distribution similar to a reference distribution. In natural images, a reference distribution not only varies from one image to another, but also within an image depending on texture. We estimate a reference distribution directly from an input image for each texture segment. Our algorithm is able to restore rich mid-frequency textures. A large-scale user study supports the conclusion that our algorithm improves the visual realism of reconstructed images compared to those of MAP estimators.

  • Paper [link],
    • @article{Cho_TPAMI2012,
      author = {Taeg Sang Cho and C. Lawrence Zitnick and Neel Joshi and Sing Bing Kang and Rick Szeliski and William T. Freeman},
      Booktitle = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
      Title = {Image restoration by matching gradient distributions},
      Year = {2012}}
  • Supplemental materials - [pdf]
  • More experimental results - [html]


This work was mostly done while the first author was an intern at MSR. This research is partially funded by NGA NEGI-1582-04-0004, by ONR-MURI Grant N00014-06-1-0734, and by gift from Microsoft, Google, Adobe. The first author is partially supported by Samsung Scholarship Foundation.