Temporal Upsampling for Image Sequences Using a Non-Local Means Algorithm

Information

• Publication Type: Master Thesis
• Workgroup(s)/Project(s):
  • Rendering and Modeling
• Date: November 2017
• Date (Start): 3. June 2014
• Date (End): 17. November 2017
• TU Wien Library:
• Diploma Examination: 17. November 2017
• First Supervisor: Michael Wimmer
• Keywords: global illumination, frame-rate upsampling, temporal coherence

Abstract

Computer-generated video sequences with a frame-rate higher than the usual 24 images per second, such as 48 or 60 frames per second, have become more popular in the respective industries, due to more visual fidelity. This, however, results in more computational costs for the same length of the video sequence.

One solution to this problem is the so-called frame-rate upsampling, which makes use of temporal and spatial coherence to approximate new frames and therefore saves computational time. Several methods have been published in this field, for the purposes of real-time rendering as well as for offline rendering algorithms.

In this thesis, two new algorithms for fame-rate upsampling are introduced. Those are targeted at high-quality computer-generated images that feature various globalillumination effects. The two new algorithms make use of a video denoising method - the non-local means algorithm - to find the appropriate pixel colors for the frame, that has to be upsampled. To find the corresponding pixels in another frame, the methods of this thesis either use existing color information or require additional data, which can be extracted from any global-illumination algorithm with minimal further computations. The proposed methods are aimed at handling reflections and refractions in the scene better than previous work.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@mastersthesis{ROEGNER-2014-TUI,
  title =      "Temporal Upsampling for Image Sequences Using a Non-Local
               Means Algorithm",
  author =     "Clemens R\"{o}gner",
  year =       "2017",
  abstract =   "Computer-generated video sequences with a frame-rate higher
               than the usual 24 images per second, such as 48 or 60 frames
               per second, have become more popular in the respective
               industries, due to more visual fidelity. This, however,
               results in more computational costs for the same length of
               the video sequence.  One solution to this problem is the
               so-called frame-rate upsampling, which makes use of temporal
               and spatial coherence to approximate new frames and
               therefore saves computational time. Several methods have
               been published in this field, for the purposes of real-time
               rendering as well as for offline rendering algorithms.   In
               this thesis, two new algorithms for fame-rate upsampling are
               introduced. Those are targeted at high-quality
               computer-generated images that feature various
               globalillumination effects. The two new algorithms make use
               of a video denoising method - the non-local means algorithm
               - to find the appropriate pixel colors for the frame, that
               has to be upsampled. To find the corresponding pixels in
               another frame, the methods of this thesis either use
               existing color information or require additional data, which
               can be extracted from any global-illumination algorithm with
               minimal further computations. The proposed methods are aimed
               at handling reflections and refractions in the scene better
               than previous work.",
  month =      nov,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  keywords =   "global illumination, frame-rate upsampling, temporal
               coherence",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2017/ROEGNER-2014-TUI/",
}