Depth Estimation Through a
Generative Model of Light Field Synthesis

Mehdi S. M. Sajjadi

Rolf Köhler

Bernhard Schölkopf

Michael Hirsch

Max-Planck Instite for Intelligent Systems
Spemanstr. 38, 72076 Tübingen, Germany

38th German Conference on Pattern Recognition (GCPR) 2016


Light field photography captures rich structural information that may facilitate a number of traditional image processing and computer vision tasks. A crucial ingredient in such endeavors is accurate depth recovery. We present a novel framework that allows the recovery of a high quality continuous depth map from light field data. To this end we propose a generative model of a light field that is fully parametrized by its corresponding depth map. The model allows for the integration of powerful regularization techniques such as a non-local means prior, facilitating accurate depth map estimation. Comparisons with previous methods show that we are able to recover faithful depth maps with much finer details. In a number of challenging real-world examples we demonstrate both the effectiveness and robustness of our approach.


Paper (pdf)Supplementary Material (pdf)Poster (pdf)

Overview of our Method

depth estimation along x, s coherence map depth estimation along y, t thresholded, combined
center sub-aperture image smooth propagation result without NLM final result with NLM

Results and Comparisons on Lightfield Images from Tao et al. (2013)

reference our result Tao et al. (2013) Wang et al. (2015) Lin et al. (2015) Sun et al. (2010) Wanner et al. (2012)

Results and Comparisons on Stanford Truck

center view our result Wanner et al. (2012) Kim et al. (2013)