IMI Interdisciplinary Mathematics InstituteCollege of Arts and Sciences

Sparse representations and digital film restoration

  • May 7, 2009


Most methods of signal processing explicitly or implicitly assume separation of useful signal or artifacts have some sparse representation. Digital film restoration provides us with very wide spectrum of signal processing problems obeying the general principles of sparse representation. At the same time, they are very different and require different approaches. Among those problems we consider:

  1. The problem of dust/dirt removal. It has some feature of Error Correcting Code.
  2. Film grain removal. It is similar to classical denoising problems.
  3. Flicker removal. The film archives community requires to leave "healthy" information unchanged.

Compare to classic requirement to maximize signal-to-noise ratio. For example, such classical methods as the Wiener filtration do not satisfy that requirement. Besides, the dimension of those problems is very high. So the sparsity is not only the way to separate data and artifact but sometimes is the only way to provide computationally efficient algorithm. All those problems are heavily dependable on the motion estimation. Which is critically important for any video processing. We also are going to address the issue of motion estimation. We will demonstrate some samples of restoration of unique materials from Library of Congress, Russian National Film Archive. Among those samples we will show a paper print (probably produced by T.Edison), Leo Tolstoy funeral, A.Schweizer plays Bach.

© Interdisciplinary Mathematics Institute | The University of South Carolina Board of Trustees | Webmaster