• Acta Photonica Sinica
  • Vol. 51, Issue 4, 0430002 (2022)
Daoquan WEI1, Huiqin WANG1、*, Ke WANG1, Zhan WANG2, and Gang ZHEN2
Author Affiliations
  • 1School of Information and Control Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China
  • 2Shaanxi Provincial Institute of Cultural Relics Protection,Xi'an 710075,China
  • show less
    DOI: 10.3788/gzxb20225104.0430002 Cite this Article
    Daoquan WEI, Huiqin WANG, Ke WANG, Zhan WANG, Gang ZHEN. Pigment Classification Method of Mural Sparse Multi-spectral Image Based on Space Spectrum Joint Feature[J]. Acta Photonica Sinica, 2022, 51(4): 0430002 Copy Citation Text show less
    Spatial feature extraction network structure
    Fig. 1. Spatial feature extraction network structure
    Spectrum feature extraction network structure
    Fig. 2. Spectrum feature extraction network structure
    Space-spectrum joint feature extraction network
    Fig. 3. Space-spectrum joint feature extraction network
    Multispectral pigment board
    Fig. 4. Multispectral pigment board
    Self-made mock murals
    Fig. 5. Self-made mock murals
    The training results of different methods under the paint board
    Fig. 6. The training results of different methods under the paint board
    Color board classification results
    Fig. 7. Color board classification results
    Classification results of different methods under self-made murals
    Fig. 8. Classification results of different methods under self-made murals
    Self-made mural classification results
    Fig. 9. Self-made mural classification results
    16-channel multispectral image of Venerable Injanta's skirt
    Fig. 10. 16-channel multispectral image of Venerable Injanta's skirt
    Partial region samples and classification results of skirts
    Fig. 11. Partial region samples and classification results of skirts
    Class numberClass nameTrainingTest
    Total3 49631 383
    1Chrome yellow2462 214
    2Orpiment2482 232
    3Garcinia2572 313
    4Head green2512 259
    5Four green2492 160
    6Head cyan2552 295
    7Cerulean2612 349
    8Four cyan2642 376
    9Lazurite2362 124
    10Crimson2512 259
    11Scarlet2492 241
    12Cinnabar2432 187
    13Ocher2472 223
    14Vermilion2392 151
    Table 1. The division of training set and test set of multispectral paint board
    ClassLSTMCNNSSJF
    Chrome yellow97.8899.1099.77
    Orpiment97.4698.2198.17
    Garcinia96.4598.2797.97
    Head green96.3799.3499.56
    Four green98.5299.6399.82
    Head cyan96.2798.5799.26
    Cerulean98.6499.5898.26
    Four cyan94.8296.9397.42
    Lazurite99.8599.8799.95
    Crimson94.5799.2599.94
    Scarlet93.5997.2497.42
    Cinnabar99.2299.7599.95
    Ocher98.2599.3799.82
    Vermilion99.8199.5899.81
    OA/%97.2898.9498.99
    Kappa×10097.0798.8698.91
    Table 2. Classification accuracy of different methods under pigment board
    LevelAbsolute measurement scaleRelative measurement scale
    1Very goodThe best in the group
    2BetterBetter than the average in the group
    3GenerallyAverage in the group
    4PoorWorse than the average in the group
    5Very badWorst in the group
    Table 3. Comparison of the scales of subjective image evaluation
    AlgorithmResult
    MDCVery bad
    SIDPoor
    SAMGenerally
    SVMGenerally
    MSCNNBetter
    LSTMBetter
    SSJFVery good
    Table 4. Evaluation results of dual stimulation injury classification method
    AlgorithmEvaluation index
    RMSEPSNRSSIM
    MDC39.8416.120.768 8
    SID28.1119.150.887 1
    SAM22.4919.150.894 4
    SVM30.9318.320.864 2
    MSCNN27.8119.250.833 9
    LSTM19.8222.190.880 2
    SSJF2.8439.060.987 4
    Table 5. Objective evaluation results of image quality
    ClassMDCSIDSAMSVMLSTMCNNSSJF
    Vermilion92.2393.3695.9699.3899.3699.3899.47
    Chrome yellow91.3799.9299.9198.9799.8699.9499.97
    Four green93.3093.2499.6198.8797.6392.88100.00
    Three green88.5398.7099.8573.4291.7093.8999.92
    Lazurite82.5094.6490.0198.2899.2399.6399.97
    Head green40.9566.3870.3497.5995.9998.1899.78
    OA/%69.3697.7597.5498.4198.5598.6899.97
    Kappa×10046.9294.4593.9697.4797.5297.7499.95
    Table 6. Classification accuracy of self-made simulated murals by different methods
    Daoquan WEI, Huiqin WANG, Ke WANG, Zhan WANG, Gang ZHEN. Pigment Classification Method of Mural Sparse Multi-spectral Image Based on Space Spectrum Joint Feature[J]. Acta Photonica Sinica, 2022, 51(4): 0430002
    Download Citation