TR2018-208

Robust Subspace Learning for Motion Deblurring in Images


    •  Lopez, O., Mansour, H., "Robust Subspace Learning for Motion Deblurring in Images," Tech. Rep. TR2018-208, Mitsubishi Electric Research Laboratories, March 2019.
      BibTeX TR2018-208 PDF
      • @techreport{Lopez2019mar,
      • author = {Lopez, Oscar and Mansour, Hassan},
      • title = {Robust Subspace Learning for Motion Deblurring in Images},
      • year = 2019,
      • month = mar,
      • url = {https://www.merl.com/publications/TR2018-208}
      • }
  • MERL Contact:
  • Research Areas:

    Computational Sensing, Machine Learning, Signal Processing

Abstract:

We developed a framework for motion deblurring that finds a low rank approximation of the sharp image patches from a collection of blurry image patches. The approach relies on the notion that each blurry patch has undergone a different type of blur compared to the other patches. As a result, the low rank approximation of the group of patches recovers a sharp image component without the misalignment artifacts associated with a rank one approximation.