Acta Optica Sinica, Volume. 44, Issue 10, 1026027(2024)

Image Synthesis of Compressive Light Field Displays with U-Net

Chen Gao1,2, Xiaodi Tan3,4,5、*, Haifeng Li6, and Xu Liu6
Author Affiliations
  • 1College of Photonic and Electronic Engineering, Fujian Normal University, Fuzhou 350117, Fujian , China
  • 2Fujian Provincial Key Laboratory of Photonics Technology, Fuzhou 350117, Fujian , China
  • 3Key Laboratory of Optoelectronic Science and Technology for Medicine of Ministry of Education, Fuzhou 350117, Fujian , China
  • 4Fujian Provincial Engineering Technology Research Center of Photoelectric Sensing Application, Fuzhou 350117, Fujian , China
  • 5Information Photonics Research Center, Fujian Normal University, Fuzhou 350117, Fujian , China
  • 6College of Optical Science and Engineering, Zhejiang University, Hangzhou 310027, Zhejiang , China
  • show less
    Figures & Tables(22)
    Principle of compressive light field display
    Schematic of projection matrix layout
    Principle of image synthesis for compressive light field displays based on ANN
    Architecture of stacked convolution neural network for compressive light field display patterns synthesis
    Architecture of U-Net for compressive light field display patterns synthesis
    Training dataset generation of target light field. (a) Target light field for training; (b) image blocks cropped from viewpoint images; (c) training dataset spliced by image blocks
    Target light field for testing
    Training and testing results of network for multiplicative-type compressive light field display patterns synthesis. (a) CNN; (b) U-Net
    Training and testing results of network for additive-type compressive light field display patterns synthesis. (a) CNN; (b) U-Net
    Training and testing results of network for polarized-type compressive light field display patterns synthesis. (a) CNN; (b) U-Net
    Training and testing results of network for hybrid-type compressive light field display patterns synthesis. (a) CNN; (b) U-Net
    Multiplicative-type compressive light field display patterns synthesized by CNN, U-Net, and iterative algorithm
    Additive-type compressive light field display patterns synthesized by CNN, U-Net, and iterative algorithm
    Polarized-type compressive light field display patterns synthesized by CNN, U-Net, and iterative algorithm
    Hybrid-type compressive light field display patterns synthesized by CNN and U-Net
    Viewpoint images of target light field and their enlargements
    Simulated reconstruction results of viewpoint images by multiplicative-type compressive light field display
    Simulated reconstruction results of viewpoint images by additive-type compressive light field display
    Simulated reconstruction results of viewpoint images by polarized-type compressive light field display
    Simulated reconstruction results of viewpoint images by hybrid-type compressive light field display
    Relationship between image quality and iterations of three basic types of compressive light field displays. (a) Multiplicative-type; (b) additive-type; (c) polarized-type
    • Table 1. Optimization objective and variable range for three basic types of compressive light field displays

      View table

      Table 1. Optimization objective and variable range for three basic types of compressive light field displays

      Type of compressive light field displaysOptimization objectiveVariable range
      Multiplicativeargminexp(Ax)E-Mbx=log t(-, 0]
      AdditiveargminAxE-Mbx=l[0, 1]
      Polarizedargminsin2(Ax)E-Mbx=θ[0, π /2]
    Tools

    Get Citation

    Copy Citation Text

    Chen Gao, Xiaodi Tan, Haifeng Li, Xu Liu. Image Synthesis of Compressive Light Field Displays with U-Net[J]. Acta Optica Sinica, 2024, 44(10): 1026027

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Physical Optics

    Received: Oct. 20, 2023

    Accepted: Dec. 1, 2023

    Published Online: May. 6, 2024

    The Author Email: Tan Xiaodi (xtan@fjnu.edu.cn)

    DOI:10.3788/AOS231683

    CSTR:32393.14.AOS231683

    Topics