• Laser & Optoelectronics Progress
  • Vol. 59, Issue 4, 0410018 (2022)
Wenteng Liu1, Huiqin Wang1、*, Ke Wang1, and Zhan Wang2
Author Affiliations
  • 1Collge of Information and Control Engineering, Xi'an University of Architecture and Technology, Xi'an , Shaanxi 710055, China
  • 2Shaanxi Institute for the Preservation of Cultural Heritage, Xi'an , Shaanxi 710075, China
  • show less
    DOI: 10.3788/LOP202259.0410018 Cite this Article Set citation alerts
    Wenteng Liu, Huiqin Wang, Ke Wang, Zhan Wang. Recognition Method of Inscription Graphics Based on Dual Features of Topology and Mesh[J]. Laser & Optoelectronics Progress, 2022, 59(4): 0410018 Copy Citation Text show less
    Topological equivalence diagram
    Fig. 1. Topological equivalence diagram
    Inscription graph topological equivalence. (a) (b) Prefix "Yuan" and its skeleton; (b) (d) prefix "Yuan" variant and its skeleton
    Fig. 2. Inscription graph topological equivalence. (a) (b) Prefix "Yuan" and its skeleton; (b) (d) prefix "Yuan" variant and its skeleton
    Schematic diagram of inscription mesh. (a) Uniform mesh; (b) Elastic mesh
    Fig. 3. Schematic diagram of inscription mesh. (a) Uniform mesh; (b) Elastic mesh
    Pixel connected area
    Fig. 4. Pixel connected area
    Four-direction correlation graph
    Fig. 5. Four-direction correlation graph
    Training technology roadmap of Bagging model
    Fig. 6. Training technology roadmap of Bagging model
    Partial inscription data
    Fig. 7. Partial inscription data
    Number of classifiers-generalization error curve
    Fig. 8. Number of classifiers-generalization error curve
    Combination relationship surface fitting diagram of base classifier
    Fig. 9. Combination relationship surface fitting diagram of base classifier
    ClassPropoesd algorithmSIFTHOGRestNet18
    Feature structure7+4×2128×n36×15×157×7×512
    11511136810025088
    21525344810025088
    31538912810025088
    41566176810025088
    51512672810025088
    Table 1. Feature dimension of different feature extraction algorithms
    Classification modelAlgorithmAverage running time /s
    BaggingPropoesd algorithm0.02
    SIFT1.53
    HOG1.31
    RestNet18RestNet180.68
    Table 2. Feature extraction algorithm model efficiency

    Five-fold

    cross validation

    BaggingRestNet18
    T-MFSIFTHOG
    Average87.9572.4166.2556.87
    189.2975.4566.5256.25
    286.1668.7564.7354.46
    387.5073.6668.7561.60
    491.0776.3467.8659.82
    585.7167.8663.3952.23
    Table 3. Accuracy of four feature extraction algorithmsunit: %
    Wenteng Liu, Huiqin Wang, Ke Wang, Zhan Wang. Recognition Method of Inscription Graphics Based on Dual Features of Topology and Mesh[J]. Laser & Optoelectronics Progress, 2022, 59(4): 0410018
    Download Citation