Soil particle size distribution by mosaic imaging and watershed analysis

A new method is presented for obtaining soil grain size distribution curves from digital images of a soil specimen. The technique offers several improvements over existing methods by adapting mosaic imaging and watershed analysis. The mosaic imaging allows the grain size distribution to be developed using a single magnification level adjusted for the finest particles in the soil. By taking many successive images on a regular grid pattern and combining them into one "mosaic'' image, a previous need for statistical corrections for grains along image boundaries is also eliminated. The watershed method for segmentation of the particles eliminates the previous requirement of having all particles in the soil detached on the viewing table. A shape adjustment factor, lambda(d), is introduced to explain the difference between sieve-based grain size distributions and digital image-based grain size distributions. As expected, lambda(d) was found to be a function of the grain shape, with values ranging from 1.10 for rounded spheres to 1.44 for elongated particles.