Determination of size distribution using neural networks

In this paper we present a novel approach to the estimation of size distributions of grains in water from images. External conditions such as the concentrations of grains in water cannot be controlled. This poses problems for local image analysis which tries to identify and measure single grains. Our approach uses global image features such as coarseness and uses a neural network to learn a mapping from these feature values to a representation of the cumulative size distribution used in the sand industry.