The effect of illumination variations on color-based wood defect classification

The spectral stability of illumination is an important aspect in color machine vision. The physical properties of lamps may change the spectrum of the illumination with time. In this work we examine the robustness of different color spaces against spectrally varying illumination in industrial wood defect inspection, when the inspection system is not adapted to the variations of illumination. The spectral changes are accomplished by digitally transforming the color temperature of illumination of wood images. We conclude that the original RGB color space is the best color space under varying illumination, the XYZ and CIELuv spaces being also very promising. An interesting observation is that features from some of the color spaces perform worse than features obtained from grey-level images. However, the importance of color information is emphasized, since RGB gives better results than grey level even under strong spectral variation of illumination.