Partial Demosaicing for Stereo Matching of CFA Images on GPU and CPU
IRTES-SET, University of Technology of Belfort-Monbeliard, Belfort, France
The Third International Conference on Advanced Communications and Computation (INFOCOMP’13), 2013
@inproceedings{zhang2013partial,
title={Partial Demosaicing for Stereo Matching of CFA Images on GPU and CPU},
author={Zhang, Naiyu and Cr{‘e}put, Jean-Charles and Wang, Hongjian and Meurie, Cyril and Ruichek, Yassine},
booktitle={INFOCOMP 2013, The Third International Conference on Advanced Communications and Computation},
pages={33–38},
year={2013}
}
This paper presents a GPU implementation of a partial demosaicing scheme that is specially designed for stereo matching of CFA image. This method consists of three main techniques keys: the adapted matching cost for CFA image, the estimated Second color component based on Hamilton’s estimate method and a robust cost aggregation window. Experiments are carried out to explore the performance for this method on GPU both at matching quality and matching efficiency, with comparison with version on CPU. The experiments on different size image pairs from Middlebury dataset show that this method can be substantially accelerated on GPU when the image size is large and has still space for improvements in performance.
December 29, 2013 by hgpu
