Neural-like thinning processing

Abstract

Neural-like approach to the thinning of binary patterns based on local thickness estimate is proposed. Thinning is considered to be a self-organizing process where a binary neuron - contour pixel value is given by the hard limiter transfer function. Input to this function is given by the pattern local thickness estimate from flexible (both in size and shape) neighborhood together with neuron threshold which ensures the preservation of the 4-connectedness and genuine end point. The final skeletons obtained within sequential, hybrid and parallel mode of operation maintain perfectly connectedness and topology. Due to the presence of the more global and structured information about pattern (involved in the local thickness estimate) the enhanced preservation of geometric properties is observed.