Resumen
Abstract. Automated image processing has great potential to aid in the classification of biological images. Many natural structures such as neurons exhibit fractal properties, and measures derived from fractal analysis are useful in differentiating neuron types. When fractal properties are not constant in all parts of the neuron, multifractal analysis may provide superior results. We applied three methods to elucidate the variation within 16 rat retinal ganglion cells: local connected fractal dimension (LCFD), mass-radius (MR) and maximum likelihood multifractal (MLM) analyses. The LCFD method suggested that some of the neurons studied are multifractal. The MR method was inconclusive due to the finite size of the cells. However, the MLM method was able to show the multifractal nature of all the samples, and to provide a superior multifractal spectrum. We conclude that the latter method warrants further attention as it may improve results in other application areas. (AU)
