Computer Graphics Laboratory ETH Zurich

ETH

Online View Sampling for Estimating Depth from Light Fields

C. Kim, K. Subr, K. Mitchell, A. Sorkine-Hornung, M. Gross

Proceedings of IEEE International Conference on Image Processing (ICIP) (Quebec City, Canada, September 27-30, 2015), pp. 1155-1159
[Abstract] [BibTeX] [PDF]

Abstract

Geometric information such as depth obtained from light fields finds more applications recently. Where and how to sample images to populate a light field is an important problem to maximize the usability of information gathered for depth reconstruction. We propose a simple analysis model for view sampling and an adaptive, online sampling algorithm tailored to light field depth reconstruction. Our model is based on the trade-off between visibility and depth resolvability for varying sampling locations, and seeks the optimal locations that best balance the two conflicting criteria.

@inproceedings{Kim15a,
    author = {Changil Kim and Kartic Subr and Kenny Mitchell and Alexander Sorkine-Hornung and Markus Gross},
    title = {Online View Sampling for Estimating Depth from Light Fields},
    booktitle = {Proceedings of IEEE International Conference on Image Processing (ICIP)},
    year = {2015},
    pages = {1155--1159},
}
[Download BibTeX]

Overview

Our algorithm constructs a preference function over the continuum of sampling locations in a cumulative manner, based on the visibility of the scene currently being captured and the accuracy of depth achievable by the depth computation algorithm in use. The best locations are determined iteratively by taking the local maxima of the current preference function, which is updated in turn by the newly sampled views.

The above fiture shows resulting depth maps computed over several iterations, where our sampling strategy in the top row is compared against the regular sampling in the bottom row. The error plots in the last column (the lower the better) show the faster convergence of ours towards lower errors.

Downloads

Download Paper
[PDF]