Skeletonization: a new application for discrete-time cellular neural networks using time-variant templates

Skeletonization is described as a novel application for discrete-time cellular neural networks (DTCNNs). The authors provide a more generalized concept of DTCNNs by use of multiple layers and time-variant templates, which are required for complex image processing tasks such as skeletonization. A special kind of time-variant template, the cyclic template, is defined. It is shown that skeletonization is achieved by a three-layer network consisting of a time-variant template layer, a linear thresholding or convolution layer, and a multiplication layer. The latter is a nonuniform processing architecture.<
>