Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 61, Issue 18, 1837014(2024)

Low-Light Image Stitching Method Based on Improved SURF

Yu Ji1,2, Peng Ding2, Nan Liu2, Zhanqiang Ru2, Zhenyao Li2, Suzhen Cheng1,2, Zhengguang Wang1,2, Jingwu Gong1,2, Zhizhen Yin2, Fei Wu2, and Helun Song1,2、*
Author Affiliations
  • 1School of Nano-Tech and Nano-Bionics, University of Science and Technology of China, Hefei 230026, Anhui,China
  • 2Suzhou Institute of Nano-Tech and Nano-Bionics, Chinese Academy of Sciences, Suzhou 215123, Jiangsu, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (15)
    Equations (15)
    References (22)
    Tools
    Paper Information
    Topics
    Figures & Tables(15)
    Feature extraction threshold and number of feature points. (a) Low illumination image; (b) normal illumination image; (c) SURF feature extraction results of normal illumination image when the threshold is 800; SURF feature extraction results of low-light image when the threshold is (d) 950, (e) 800, (f) 650, (g) 500, (h) 350, (i) 200, and (j) 50

    Fig. 1. Feature extraction threshold and number of feature points. (a) Low illumination image; (b) normal illumination image; (c) SURF feature extraction results of normal illumination image when the threshold is 800; SURF feature extraction results of low-light image when the threshold is (d) 950, (e) 800, (f) 650, (g) 500, (h) 350, (i) 200, and (j) 50

    Download full size
    Quantitative analysis of feature thresholds and number of feature points

    Fig. 2. Quantitative analysis of feature thresholds and number of feature points

    Download full size
    Illumination and edge information of the image. (a) (d) Input images; (b) (e) luminance estimation images; (c) (f) edge extraction images

    Fig. 3. Illumination and edge information of the image. (a) (d) Input images; (b) (e) luminance estimation images; (c) (f) edge extraction images

    Download full size
    Calculate ESR image. (a) (d) Binarized input images; (b) (e) binarized edge images; (c) (f) ESR images

    Fig. 4. Calculate ESR image. (a) (d) Binarized input images; (b) (e) binarized edge images; (c) (f) ESR images

    Download full size
    RMSE of Gaussian filter kernel and mean filter kernel

    Fig. 5. RMSE of Gaussian filter kernel and mean filter kernel

    Download full size
    Improved SURF low-light image stitching process

    Fig. 6. Improved SURF low-light image stitching process

    Download full size
    Comparison of outdoor low-light image enhancement effects and enlarged details. (a) Original image; (b) SSR algorithm; (c) MSR algorithm; (d) MSRCR algorithm; (e) image enhancement algorithm in GIMP; (f) proposed algorithm

    Fig. 7. Comparison of outdoor low-light image enhancement effects and enlarged details. (a) Original image; (b) SSR algorithm; (c) MSR algorithm; (d) MSRCR algorithm; (e) image enhancement algorithm in GIMP; (f) proposed algorithm

    Download full size
    Comparison of indoor low-light image enhancement effects and enlarged details. (a) Original image; (b) SSR algorithm; (c) MSR algorithm; (d) MSRCR algorithm; (e) image enhancement algorithm in GIMP; (f) proposed algorithm

    Fig. 8. Comparison of indoor low-light image enhancement effects and enlarged details. (a) Original image; (b) SSR algorithm; (c) MSR algorithm; (d) MSRCR algorithm; (e) image enhancement algorithm in GIMP; (f) proposed algorithm

    Download full size
    Feature extraction under different algorithms and thresholds. (a) Original image; (b) SURF algorithm, Hessian threshold is 800; (c) SURF algorithm, Hessian threshold is 400; (d) SURF algorithm, Hessian threshold is 100; (e) SIFT algorithm, contrast threshold is 0.06; (f) SIFT algorithm, contrast threshold is 0.04; (g) SIFT algorithm, contrast threshold is 0.02; (h) proposed algorithm

    Fig. 9. Feature extraction under different algorithms and thresholds. (a) Original image; (b) SURF algorithm, Hessian threshold is 800; (c) SURF algorithm, Hessian threshold is 400; (d) SURF algorithm, Hessian threshold is 100; (e) SIFT algorithm, contrast threshold is 0.06; (f) SIFT algorithm, contrast threshold is 0.04; (g) SIFT algorithm, contrast threshold is 0.02; (h) proposed algorithm

    Download full size
    Feature matching and RANSAC optimization based on fast filtering. (a) Feature rough matching results based on fast filtering; (b) feature matching results after RANSAC optimization

    Fig. 10. Feature matching and RANSAC optimization based on fast filtering. (a) Feature rough matching results based on fast filtering; (b) feature matching results after RANSAC optimization

    Download full size
    Splicing effect of some low-light images in the data set

    Fig. 11. Splicing effect of some low-light images in the data set

    Download full size
    Comparison of low-light image stitching outcomes at varying overlap rates

    Fig. 12. Comparison of low-light image stitching outcomes at varying overlap rates

    Download full size

    • Table 1. Comparison of evaluation indicators of different algorithms

      View table

      Table 1. Comparison of evaluation indicators of different algorithms

      MethodENSDRMSEPSNRSSIMGMSDTMQI
      SSR13.91229.08935.18017.2050.8160.0730.842
      MSR14.48820.60444.33115.1970.7990.1050.818
      MSRCR14.13535.59961.96312.2880.7840.0610.762
      GIMP11.86137.91617.92423.0620.8800.0560.908
      Proposed16.03651.09415.81724.3810.8310.0420.945

    • Table 2. Feature extraction results under different algorithms and different thresholds

      View table

      Table 2. Feature extraction results under different algorithms and different thresholds

      MethodImage 1Image 2
      Low illuminationNormal illuminationOverlapping rate /%Low illuminationNormal illuminationOverlapping rate /%
      SURF（Hessian is 800）276124.4419714.2
      SURF（Hessian is 400）10110749.4134130310.3
      SURF（Hessian is 100）637245825.9603198930.3
      SIFT（contrast is 0.06）264595.7288143.4
      SIFT（contrast is 0.04）9576512.410898910.9
      SIFT（contrast is 0.02）565125145.2525115245.6
      Proposed1189157475.5751119662.8

    • Table 3. Matching algorithm performance comparison

      View table

      Table 3. Matching algorithm performance comparison

      MethodRough matchCorrect matchCMR /%Total time /sSignal match time /ms
      SURF8961680.222.50
      SIFT8463750.182.10
      BRIEF（brute force）495341680.480.97
      BRIEF（random KD tree）471398630.390.83
      BRIEF（proposed）289276950.160.55
    Tools

    Get Citation

    Copy Citation Text

    Yu Ji, Peng Ding, Nan Liu, Zhanqiang Ru, Zhenyao Li, Suzhen Cheng, Zhengguang Wang, Jingwu Gong, Zhizhen Yin, Fei Wu, Helun Song. Low-Light Image Stitching Method Based on Improved SURF[J]. Laser & Optoelectronics Progress, 2024, 61(18): 1837014

    Download Citation

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

    Category: Digital Image Processing

    Received: Jan. 8, 2024

    Accepted: Feb. 18, 2024

    Published Online: Sep. 14, 2024

    The Author Email: Helun Song (hlsong2008@sinano.ac.cn)

    DOI:10.3788/LOP240470

    CSTR:32186.14.LOP240470

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology