zur Startseite

Masterarbeit

CNN-based plenoptic Image Super-resolution
Betreuer M.Eng. Trung Hieu Tran
Prüfer Prof. Dr.-Ing. Sven Simon
Ende
Beschreibung

Light-field (LF) cameras provides an efficient way to capture both spatial and angular information of the scene. These rich contents encoded in LF image brings a great benefit to many applications such as computer vision, free-viewpoint TV, refocused imaging, ... Among different types of LF acquisitions, plenoptic camera ( such as from Lytro, Raytrix, ... ) is the most convenient one but is also suffered from the low spatial/angular resolution problem.

In this master thesis / Diplomarbeit an efficient framework for super-resolution of plenoptic LF image will be studied and implemented on GPU computing platform. As an essential task, the correspondence problem need to be solved first for extracting disparity map. Then compiling this information with original views to reconstruct a higher resolution view as well as synthesize novel views. Depend on processing tasks, Matlab coding would be involved as a rapid way to assess the algorithms before the implementation on GPU. The duration of the thesis will be 6 months (full-time).

Prerequisites:

  • Experience with Matlab/Python
  • Experience with deep learning frameworks (PyTorch/Tensorflow)

Contact:

Trung Hieu Tran (trung.hieu.tran@ipvs.uni-stuttgart.de)

Download PDF