Automatic thresholding for digital hemispherical photography

This paper proposes an automatic thresholding method for the discrimination of sky and canopy elements in color hemispherical photographs taken with a digital camera (Nikon Coolpix 950). The exposures for photography were principally determined on the basis of zenith luminance. DIFphoto, which is diffuse transmittance calculated from the hemispherical photographs, was related to DIFsensor, which is diffuse transmittance measured directly with a photosynthetic photon flux density sensor. First, the thresholds for calculation of DIFphoto were manually assessed in the photographs to obtain the best match with DIFsensor. At the lower pixel-value level in the pixel histograms from the photographs, L-shaped curves were always recognized, and the threshold occurred at the point with the maximum curvature. Second, an automatic thresholding algorithm, taking into account the position of the thresholds, was computerized. Third, the relationships between DIFphoto and DIFsensor were field-tested across a wide range of light conditions. The method was effective in a planted coniferous forest and a natural broad-leaved forest and under overcast, twilit, and sunny sky conditions. The coefficients of determination between DIFsensor and DIFphoto were greater than 0.99. However, DIFphoto taken with Auto-Exposure was overestimated under dense canopy.