A Skeletonizing Reconfigurable Self-Organizing Model: Validation Through Text Recognition

摘要

Self Organizing Maps are able to develop topology preserving classifiers. In this work we propose a Reconfigurable Self Organizing Model, which combines this property with others related with the generation of sub-graphs of the Delaunay-triangulation, the possibility of generating elastic approximations and the capacity to reconfigure the models topological structure in a data driven way. These properties allow us to apply the model to the extraction of linear structures from one-dimensional curves and from two-dimensional figures (which can be dense or not). Skeletonization and recognition of machine printed text and handwritten numerals serve as a validation domain.