Документ взят из кэша поисковой машины. Адрес оригинального документа : http://graphics.cs.msu.ru/en/node/1026
Дата изменения: Sun Apr 10 00:38:14 2016
Дата индексирования: Sun Apr 10 00:38:14 2016
Кодировка: UTF-8
High quality stereo-to-multiview generation | Graphics and Media Lab

High quality stereo-to-multiview generation

Contact person: Dmitriy S. Vatolin (dmitriy@graphics.cs.msu.ru)

picture

Main problems:

  • dealing with fuzzy edges
  • filling occlusion areas when extrapolating new views

Scheme of our multiview generation process

picture

Dealing with fuzzy edges

When it is necessary:

  • Edges with very complicated structure (hair, fur and etc.)
  • Some pixels of source image may be a mixture of background and foreground due to
    • optical blur
    • motion blur
    • sub-pixel object shifts
  • Solution: use Matting to decompose source image to background image, foreground image and transparency map

picture

Filling occlusion areas

When extrapolating new views source images don't contain all necessary information to fill occlusions on edges of depth map
Solution:

  • Temporal inpainting ? take missing information from other frames
  • Spatial inpainting ? use current frame to generate missing information

picture picture picture

Spatial inpainting techniques comparison

picture

picture

Team

    Acknowledgements

    This work is partially supported by the Intel/Cisco Video-Aware Wireless Network (VAWN) Program