Machine Learning and Intelligent Communications. Second International Conference, MLICOM 2017, Weihai, China, August 5-6, 2017, Proceedings, Part I

Research Article

Global Depth Refinement Based on Patches

Download
177 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-73564-1_42,
        author={Xu Huang and Yanfeng Zhang and Gang Zhou and Lu Liu and Gangshan Cai},
        title={Global Depth Refinement Based on Patches},
        proceedings={Machine Learning and Intelligent Communications. Second International Conference, MLICOM 2017, Weihai, China, August 5-6, 2017, Proceedings, Part I},
        proceedings_a={MLICOM},
        year={2018},
        month={2},
        keywords={1D label 3D label Fronto-parallel bias Patch Global optimization},
        doi={10.1007/978-3-319-73564-1_42}
    }
    
  • Xu Huang
    Yanfeng Zhang
    Gang Zhou
    Lu Liu
    Gangshan Cai
    Year: 2018
    Global Depth Refinement Based on Patches
    MLICOM
    Springer
    DOI: 10.1007/978-3-319-73564-1_42
Xu Huang1,*, Yanfeng Zhang2, Gang Zhou1, Lu Liu1, Gangshan Cai1
  • 1: Wuhan Engineering Science and Technology Institute
  • 2: Wuhan University
*Contact email: huangxu.chess@163.com

Abstract

Current stereo matching methods can be divided into 1D label algorithms and 3D label algorithms. 1D label algorithms are simple and fast, but they can’t aovid fronto-parallel bias. 3D label algorithms can solve fronto-parallel bias. However, they are very time-consuming. In order to avoid fronto-parallel bias efficiently, this paper introduces a new global depth refinement based on patches. The method transforms the depth optimization problem into a quadratic function computation, which has a low time complexity. Experiments on Motorcycle imagery and Wuhan university imagery verify the correctness and the effectiveness of the proposed method.