Feature grid neural networks for curve partitioning

Presents a neural network method for partitioning image curves into perceptual entities called generic curve segments (GCSs). GCSs are perceptual classes of primitive curve objects, which are qualitative descriptors for grouping curve shapes. The success of GCS classification and curve grouping relies on correctly locating curve partitioning points (CPPs), i.e. points from where the curves are broken down into GCSs. In this paper, a feature grid model is presented based on perceptual organization principles for establishing the training set and the input structure of the neural networks. The effectiveness of the method is demonstrated by experimental results