• Laser & Optoelectronics Progress
  • Vol. 58, Issue 22, 2210010 (2021)
Zhongzhi Tang, Bing Yan*, Yan Huang**, Chunrong Hua, and Dong Zheng
Author Affiliations
  • School of Mechanical Engineering, Southwest Jiaotong University, Chengdu, Sichuan 610031, China
  • show less
    DOI: 10.3788/LOP202158.2210010 Cite this Article Set citation alerts
    Zhongzhi Tang, Bing Yan, Yan Huang, Chunrong Hua, Dong Zheng. Modified SIFT Algorithm for Image Stereo Matching Based on Bidirectional Pre-Screening[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2210010 Copy Citation Text show less
    Simulation of pixel filter
    Fig. 1. Simulation of pixel filter
    Schematic of 8-neighborhood ring template
    Fig. 2. Schematic of 8-neighborhood ring template
    Processing flow of improved SIFT algorithm
    Fig. 3. Processing flow of improved SIFT algorithm
    Test results of images. (a) Test images from different angles; (b) displacement distribution in x direction; (c) displacement distribution in y direction; (d) displacement distribution in main direction
    Fig. 4. Test results of images. (a) Test images from different angles; (b) displacement distribution in x direction; (c) displacement distribution in y direction; (d) displacement distribution in main direction
    Test images. (a)(b)(c)(d) Different perspectives; (e) different illumination; (f) different degrees of ambiguity; (g) different degrees of compression
    Fig. 5. Test images. (a)(b)(c)(d) Different perspectives; (e) different illumination; (f) different degrees of ambiguity; (g) different degrees of compression
    Feature point matching results of three algorithms. (a) SIFT+RANSAC algorithm; (b) SP-SIFT+RANSAC algorithm; (c) MSIFT+MRANSAC algorithm
    Fig. 6. Feature point matching results of three algorithms. (a) SIFT+RANSAC algorithm; (b) SP-SIFT+RANSAC algorithm; (c) MSIFT+MRANSAC algorithm
    ImageSize /(pixel×pixel)SIFTSP-SIFTMSIFT
    NumberTime /sNumberTime /sNumberTime /s
    Road 11241×37620612.0817411.7519811.81
    Road 21241×37633342.1526801.8332561.97
    Teddy450×3757061.165570.936910.97
    Plant960×54032812.2229401.9432452.09
    Car900×60025182.2121401.7624762.01
    Bike1000×70034072.7629022.1633552.43
    Church800×64055612.9149362.2655112.40
    Table 1. Feature point extraction results of each algorithm
    IndexAlgorithmImage
    Road 1Road 2TeddyPlantCarBikeChurch
    Number oftotal matchesSIFT+RANSAC140321174872474151413404094
    SP-SIFT+RANSAC11731675391219712809263624
    MSIFT+MRANSAC134620504722440146613094046
    Number ofcorrect matchesSIFT+RANSAC444439200190111847032578
    SP-SIFT+RANSAC38936716116309494672174
    MSIFT+MRANSAC451448207199211307802609
    P /%SIFT+RANSAC98.2699.1097.0999.95100.0099.7299.96
    SP-SIFT+RANSAC98.9898.9299.3899.94100.0099.7999.86
    MSIFT+MRANSAC98.4599.1299.52100.0099.91100.00100.00
    R /%SIFT+RANSAC93.5897.5687.7292.6099.9285.5296.45
    SP-SIFT+RANSAC92.4089.9592.8490.56100.0093.7899.77
    MSIFT+MRANSAC97.5894.3293.0598.7699.9199.87100.00
    F /%SIFT+RANSAC95.8698.3292.1796.1399.9692.0898.17
    SP-SIFT+RANSAC95.5894.2296.0095.02100.0096.6999.82
    MSIFT+MRANSAC98.0196.6696.1899.3899.9199.94100.00
    Error /pixelSIFT+RANSAC1.20570.88561.36410.38230.35140.50240.4077
    SP-SIFT+RANSAC0.90621.05731.21930.36710.34420.49680.4192
    MSIFT+MRANSAC0.87520.98181.17660.39000.34620.53630.4139
    Table 2. Comprehensive performance comparison of each algorithm
    ImageSIFT+RANSACSP-SIFT+RANSACMSIFT+MRANSAC
    ExtractionMatchingTotal timeExtractionMatchingTotal timeExtractionMatchingTotal time
    Road 14.165.9210.083.504.808.303.625.198.81
    Road 24.308.1312.433.666.309.963.947.4111.35
    Teddy2.322.564.881.862.123.981.942.394.33
    Plant4.448.3512.793.886.9510.834.187.3711.55
    Car4.426.1310.553.525.188.704.025.659.67
    Bike5.527.7813.304.325.419.734.866.7611.62
    Church5.8214.7120.534.5212.1116.634.8013.3118.11
    Table 3. Time consuming comparison of each algorithm unit: s
    Zhongzhi Tang, Bing Yan, Yan Huang, Chunrong Hua, Dong Zheng. Modified SIFT Algorithm for Image Stereo Matching Based on Bidirectional Pre-Screening[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2210010
    Download Citation