Light Field Segmentation Using a Ray-Based Graph Structure

Abstract

• We are interested in interactive light field segmentation from a reference view. • Challenge for light fields: the amount of data to process in order to edit all the views simultaneously. • In particular, the running time for graph-based approaches, such as graph-cut, increases greatly with the size of the input graph. • We assume a depth map to be known for each view. Our Method Building The Graph • Previous work focuses on representations with one graph node per ray [1]. • Observation: many rays of the light field mainly describe the same content. • The redundancy is captured by depth estimation. • We use a single node to represent several rays coming from the same scene point (ray bundles), according to an estimated measure: ⎧⎨ ⎩ [xi + (si − sj)D(si, ti, xi, yi)] = xj [xj + (sj − si)D(sj, tj, xj, yj)] = xi

DOI: 10.1007/978-3-319-46478-7_3

Cite this paper

@inproceedings{Hog2016LightFS, title={Light Field Segmentation Using a Ray-Based Graph Structure}, author={Matthieu Hog and Neus Sabater and Christine Guillemot}, booktitle={ECCV}, year={2016} }