Dynamic Super-Rays for Efficient Light Field Video Processing

Matthieu Hog (1, 2) , Neus Sabater (2) , Christine Guillemot (1)

1 Sirocco - Analysis representation, compression and communication of visual data
2 Technicolor R & I [Cesson Sévigné]

One challenging problem with light field video editing is the dreadful volume of data to process. Image and video processing frequently rely on over-segmentation methods to reduce the computational burden of subsequent editing tasks. In this paper, we present the first approach for video light field over-segmentation called dynamic super-rays, which can be seen as temporally and angularly consistent superpixels. Our algorithm is memory efficient and fast. The results show timing close to editing time by lever-aging GPU computational power.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Matthieu Hog, Neus Sabater, Christine Guillemot. Dynamic Super-Rays for Efficient Light Field Video Processing. BMVC 2018 - 29th British Machine Vision Conference, Sep 2018, Newcastle upon Tyne, United Kingdom. pp.1-12. ⟨hal-01649342v2⟩