Acta Optica Sinica, Volume. 44, Issue 9, 0915002(2024)

Unsupervised Learning Based Image Registration of Wind Tunnel Pressure Sensitive Paint Image

Kang Liu1,2,3, Xiongwei Sun1,2,3、*, Hailiang Shi1,2,3、**, Xianhua Wang1,2,3, Hanhan Ye2,3, Chen Cheng1,2,3, Feng Zhu2,3, and Shichao Wu2,3
Author Affiliations
  • 1University of Science and Technology of China, Hefei 230026, Anhui, China
  • 2Anhui Institute of Optics and Fine Mechanics, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, Anhui, China
  • 3Key Laboratory of General Optical Calibration and Characterization Technology, Chinese Academy of Sciences, Hefei 230031, Anhui, China
  • show less
    Figures & Tables(17)
    Flow of PSP experiment data processing
    Structure of PIR-Net
    Structure of convolutional networks
    PSP images of shell part. (a) Windy image; (b) windless image; (c) superimposed display image
    PSP images of sheet part. (a) Windy image; (b) windless image; (c) superimposed display image
    Registration error images of shell part. (a) Error images of windy image; (b) error images of Feature Matching registration; (c) error images of MI-Spline registration; (d) error images of Voxelmorph registration; (e) error images of CycleMorph registration; (f) error images of BIRGU-Net registration; (g) error images of LRN registration; (h) error images of PIR-Net registration
    Registration error images of sheet part. (a) Error images of windy image; (b) error images of Feature Matching registration; (c) error images of MI-Spline registration; (d) error images of Voxelmorph registration; (e) error images of CycleMorph registration; (f) error images of BIRGU-Net registration; (g) error images of LRN registration; (h) error images of PIR-Net registration
    Distribution of key points of shell part
    Distribution of key points of sheet part
    Maximum error positions and average error values of shell part registration results. (a) Error images of windy image; (b) error images of Feature Matching registration; (c) error images of MI-Spline registration; (d) error images of Voxelmorph registration; (e) error images of CycleMorph registration; (f) error images of BIRGU-Net registration; (g) error images of LRN registration; (h) error images of PIR-Net registration
    Maximum error positions and average error values of sheet part registration results. (a) Error images of windy image; (b) error images of Feature Matching registration; (c) error images of MI-Spline registration; (d) error images of Voxelmorph registration; (e) error images of CycleMorph registration; (f) error images of BIRGU-Net registration; (g) error images of LRN registration; (h) error images of PIR-Net registration
    • Table 1. Comparison of registration accuracy of shell part

      View table

      Table 1. Comparison of registration accuracy of shell part

      MethodOriginalFeature MatchingMI-SplineVoxelmorphCycleMorphBIRGU-NetLRNPIR-Net
      RMSE0.13380.15440.10430.07210.06460.07150.06240.0505
      NCC0.17220.14510.11960.33480.34830.24430.35120.4087
    • Table 2. Comparison of registration accuracy of sheet part

      View table

      Table 2. Comparison of registration accuracy of sheet part

      MethodOriginalFeature MatchingMI-SplineVoxelmorphCycleMorphBIRGU-NetLRNPIR-Net
      RMSE0.17840.15740.14230.16960.16350.16810.11820.0479
      NCC0.02430.02370.03980.02690.02610.02550.03920.0596
    • Table 3. TRE comparison of shell part

      View table

      Table 3. TRE comparison of shell part

      MethodTRE /pixel
      Point 1Point 2Point 3Point 4Point 5Point 6
      Original7.075.094.129.218.068.54
      Feature Matching4.474.242.825.835.655.09
      MI-Spline3.162.824.243.606.083.16
      Voxelmorph1.411.211.638.066.327.07
      CycleMorph1.201.071.415.094.565.04
      BIRGU-Net1.482.111.669.057.078.06
      LRN1.212.232.138.243.047.28
      PIR-Net1.130.711.101.331.411.27
    • Table 4. TRE comparison of sheet part

      View table

      Table 4. TRE comparison of sheet part

      MethodTRE /pixel
      Point 1Point 2Point 3Point 4Point 5Point 6
      Original6.084.123.6011.1810.2919.10
      Feature Matching3.168.0615.268.6111.0415.03
      MI-Spline13.3415.5219.1318.8614.319.05
      Voxelmorph7.145.384.1213.1512.2118.02
      CycleMorph6.085.044.5610.048.6016.12
      BIRGU-Net5.394.124.4710.199.2114.03
      LRN4.313.612.828.544.246.71
      PIR-Net2.241.701.412.232.112.62
    • Table 5. Comparison of registration time for different methods

      View table

      Table 5. Comparison of registration time for different methods

      MethodFeature MatchingMI-SplineVoxelmorphCycleMorphBIRGU-NetLRNPIR-Net
      Registration time for shell part /s1.767540.410.390.420.450.48
      Registration time for sheet part /s1.757610.430.420.430.470.51
    • Table 6. Impact of the number of cascaded convolutional models on registration accuracy and time

      View table

      Table 6. Impact of the number of cascaded convolutional models on registration accuracy and time

      Number of cascading CNN modelsShell partSheet part
      RMSENCCRegistration time /sRMSENCCRegistration time /s
      1 scale (1)0.07410.33700.410.17860.02370.43
      2 scale (1/2, 1)0.06580.34760.450.11680.03860.48
      3 scale (1/4, 1/2, 1)0.05050.40870.480.04790.05960.52
      4 scale (1/8, 1/4, 1/2, 1)0.04960.40200.510.04830.05880.54
    Tools

    Get Citation

    Copy Citation Text

    Kang Liu, Xiongwei Sun, Hailiang Shi, Xianhua Wang, Hanhan Ye, Chen Cheng, Feng Zhu, Shichao Wu. Unsupervised Learning Based Image Registration of Wind Tunnel Pressure Sensitive Paint Image[J]. Acta Optica Sinica, 2024, 44(9): 0915002

    Download Citation

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

    Category: Machine Vision

    Received: Dec. 4, 2023

    Accepted: Feb. 23, 2024

    Published Online: May. 15, 2024

    The Author Email: Xiongwei Sun (xiongweisun@163.com), Hailiang Shi (hlshi@aiofm.ac.cn)

    DOI:10.3788/AOS231885

    CSTR:32393.14.AOS231885

    Topics