• Laser & Optoelectronics Progress
  • Vol. 55, Issue 12, 121007 (2018)
Mei Yang*, Zefu Tan, Li Cai, and Xue Yao
Author Affiliations
  • Key Laboratory of Signal and Information Prcessing, Chongqing Three Gorges University, Chongqing 404000, China
  • show less
    DOI: 10.3788/LOP55.121007 Cite this Article Set citation alerts
    Mei Yang, Zefu Tan, Li Cai, Xue Yao. Illumination Compensation for Face Images Based on Anisotropic Retinex[J]. Laser & Optoelectronics Progress, 2018, 55(12): 121007 Copy Citation Text show less
    Pixel mean of image row and column. (a)(d) Original image; (b)(e) pixel mean in column; (c)(f) pixel mean in row
    Fig. 1. Pixel mean of image row and column. (a)(d) Original image; (b)(e) pixel mean in column; (c)(f) pixel mean in row
    Schematic of curvature
    Fig. 2. Schematic of curvature
    Face false edge extraction. (a) Original image; (b) face edge; (c) edge of a Cartesian coordinate system; (d) false edge of largest connected domain and no symmetry
    Fig. 3. Face false edge extraction. (a) Original image; (b) face edge; (c) edge of a Cartesian coordinate system; (d) false edge of largest connected domain and no symmetry
    Face false edge of small slope. (a) t=0; (b) t=0.1; (c) t=0.2; (d) t=0.3
    Fig. 4. Face false edge of small slope. (a) t=0; (b) t=0.1; (c) t=0.2; (d) t=0.3
    False edge with a small curvature and opposite direction of the light source
    Fig. 5. False edge with a small curvature and opposite direction of the light source
    Edge markup results need to be enhanced before and after improvement. (a) Original image; (b)-(f) improved edge markup results with m is 0.01、0.03、0.05、0.07、0.1, respectively; (g)-(k) corresponding edge markup results before improvement
    Fig. 6. Edge markup results need to be enhanced before and after improvement. (a) Original image; (b)-(f) improved edge markup results with m is 0.01、0.03、0.05、0.07、0.1, respectively; (g)-(k) corresponding edge markup results before improvement
    Process illumination compensation of proposed method
    Fig. 7. Process illumination compensation of proposed method
    Illumination angle is 0°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Fig. 8. Illumination angle is 0°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Illumination angle is 15°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Fig. 9. Illumination angle is 15°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Illumination angle is -20°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Fig. 10. Illumination angle is -20°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Illumination angle is +65°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Fig. 11. Illumination angle is +65°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Illumination angle is +90°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Fig. 12. Illumination angle is +90°. (a) Original image; (b) improved MSR; (c) PCNN; (d) adaptive Gamma correction; (e) proposed method
    Experiment of MCU PIE face database. (a) Original images; (b) improved MSR; (c) PCNN; (d) proposed method
    Fig. 13. Experiment of MCU PIE face database. (a) Original images; (b) improved MSR; (c) PCNN; (d) proposed method
    Face recognition rate of different subsets
    Fig. 14. Face recognition rate of different subsets
    Face misjudgment rate of different subsets
    Fig. 15. Face misjudgment rate of different subsets
    No.AlgorithmMeanRMSDLSDCSFNRS
    1Original73.2158.6965.137.15
    Improved MSR191.4136.93184.764.05
    PCNN166.5547.74155.527.46
    Adaptive Gamma correction119.1872.05102.2409.84
    Proposed method195.3237.66189.458.13
    2Original44.5559.7339.983.97
    Improved MSR157.7949.49150.015.44
    PCNN133.7662.44125.415.52
    Adaptive Gamma correction122.6070.20119.179.25
    Proposed method177.4460.15158.5912.10
    3Original42.9161.4838.755.90
    Improved MSR153.0245.81144.775.43
    PCNN130.6162.17122.165.62
    Adaptive Gamma correction130.4070.72121.079.2
    Proposed method162.1552.70143.0212.81
    4Original46.1045.8741.113.70
    Improved MSR164.0149.48155.585.15
    PCNN150.4357.92141.635.99
    Adaptive Gamma correction128.5172.32115.976.97
    Proposed method196.9146.08179.7710.18
    5Original19.3235.9423.275.57
    Improved MSR131.9350.40121.797.48
    PCNN102.1750.8191.788.14
    Adaptive Gamma correction132.8270.30119.259.41
    Proposed method144.6450.81130.329.97
    Table 1. Evaluation values of test image
    Mei Yang, Zefu Tan, Li Cai, Xue Yao. Illumination Compensation for Face Images Based on Anisotropic Retinex[J]. Laser & Optoelectronics Progress, 2018, 55(12): 121007
    Download Citation