PARTICLE-SHAPE CHARACTERIZATION USING IMAGE-ANALYSIS AND NEURAL NETWORKS

An image analysis method is presented which effectively describes the shape of a convex or concave particle. The method uses the Fourier des criptors evaluated from the Fourier series expansion of the angular be nd of the periphery of a particle as a function of its are length. The Fourier descriptors are then used as inputs to unsupervised or superv ised artificial neural networks to cluster and classify particles acco rding to their shape. A number describing the class of a single partic le or the average class of a population of particles can therefore be deduced to characterize them.