• Laser & Optoelectronics Progress
  • Vol. 56, Issue 14, 141008 (2019)
Honghao Zhou1、2, Weining Yi2, Lili Du2、*, and Yanli Qiao1、2、**
Author Affiliations
  • 1 School of Environmental Science and Optoelectronic Technology, University of Science and Technology of China, Hefei, Anhui 230031, China
  • 2 Key Laboratory of Optical Calibration and Characterization, Chinese Academy of Sciences, Hefei, Anhui 230031, China
  • show less
    DOI: 10.3788/LOP56.141008 Cite this Article Set citation alerts
    Honghao Zhou, Weining Yi, Lili Du, Yanli Qiao. Convolutional Neural Network-Based Dimensionality Reduction Method for Image Feature Descriptors Extracted Using Scale-Invariant Feature Transform[J]. Laser & Optoelectronics Progress, 2019, 56(14): 141008 Copy Citation Text show less
    Reshapeschematic of SIFT feature descriptor
    Fig. 1. Reshapeschematic of SIFT feature descriptor
    SIFT feature descriptors after changing shape. (a) Feature descriptor; (b) feature descriptor matched with Fig. 2(a); (c) feature descriptor unmatched with Fig. 2(a)
    Fig. 2. SIFT feature descriptors after changing shape. (a) Feature descriptor; (b) feature descriptor matched with Fig. 2(a); (c) feature descriptor unmatched with Fig. 2(a)
    Diagram of convolutional neural network for SIFT feature descriptor dimensionality reduction
    Fig. 3. Diagram of convolutional neural network for SIFT feature descriptor dimensionality reduction
    Partial atlases used to test matching performance of feature descriptor
    Fig. 4. Partial atlases used to test matching performance of feature descriptor
    Comparison of partial matching results between CNN-SIFT feature descriptor and SIFT feature descriptor in affine transformation
    Fig. 5. Comparison of partial matching results between CNN-SIFT feature descriptor and SIFT feature descriptor in affine transformation
    Matching performance of feature descriptors in rotation and scale transformations
    Fig. 6. Matching performance of feature descriptors in rotation and scale transformations
    Matching performance of feature descriptors in viewpoint transformation
    Fig. 7. Matching performance of feature descriptors in viewpoint transformation
    Matching performance of feature descriptors in light transformation
    Fig. 8. Matching performance of feature descriptors in light transformation
    Average matching time of feature descriptors in Webcam dataset
    Fig. 9. Average matching time of feature descriptors in Webcam dataset
    Layer12345
    Input size16×88×44×21×1281×96
    Filter size3×33×33×3--
    Number of output channel32486411
    Max-pooling size2×22×22×2--
    Nonlinearitytanhtanhtanhtanhtanh
    Out size8×44×22×11×96N
    Table 1. Structural parameters of network
    Honghao Zhou, Weining Yi, Lili Du, Yanli Qiao. Convolutional Neural Network-Based Dimensionality Reduction Method for Image Feature Descriptors Extracted Using Scale-Invariant Feature Transform[J]. Laser & Optoelectronics Progress, 2019, 56(14): 141008
    Download Citation