In this paper, we present a novel three dimensional segmentation method for a set of two dimensional sequential images such as fMRI images. The method is based on Pulse-Coupled Neural Network(PCNN), which is originally proposed for a possible explanation of the synchronous burst on cat's visual cortex by Eckhorn et. al. PCNN has a smoothing process in itself and it doesn't require a learning process. With these properties, PCNN gives good results for image processing such as edge detection, two dimensional segmentation, extracting invariant feature as a time signature, which is invariant under small distortion, rotation, scale transformation. PCNN consists of three parts, feeding part including the external stimulus, linking part, pulse generator. With extending PCNN to three dimensional one with spherical receptive fields for feeding and linking parts, for a given set of two dimensional sequential images corresponding to an external stimulus, the extended three dimensional PCNN can segment the set of two dimensional sequential images into each three dimensional region depending on the averaged intensity in the region automatically without any kinds of learning process. Some three dimensional segmentation results are also given in order to show the effectiveness of our three dimensional segmentation method.

View full abstract