4D light field reconstruction and scene depth estimation from plenoptic camera raw images
Shan, Xu
Shan, Xu
Loading...
Repository DOI
Publication Date
2016-05-16
Type
Thesis
Downloads
Citation
Abstract
Microlens-array-based light-field cameras (also known as plenoptic cameras), which are capable of recording both angular and spatial information of light, are already commercially available as consumer commodities. Due to the fact that we are still using the 2D planar sensor to record the 4D light field, extra optical design and post-processing considerations are required to reconstruct a high-quality light field. In this thesis, three research problems are investigated: light-field camera simulation, light field reconstruction, and depth from light field. First, we propose a framework for computational camera simulation utilising computer graphic rendering software. Using this simulation framework, we analyse image formation in the microlens-array-based light-field camera. Second, we present a reconstruction algorithm that is able to reconstruct the 4D light field from a portable plenoptic camera without the need for calibration images. Our quality assessment shows that our proposed approach achieves comparable results to the state-of-the-art algorithms that require calibration images. Third, we present the theoretical analysis of depth resolvability for the microlens-array-based light-field camera. A simple and efficient implementation of depth from light field algorithm, in particular, a novel and effective refinement scheme for Lytro camera is proposed. We demonstrate that our approach outperforms the state-of-the-art algorithms and commercial software.
Funder
Publisher
Publisher DOI
Rights
Attribution-NonCommercial-NoDerivs 3.0 Ireland