Rodrigo de Oliveira Plotze,  Mário Augusto Pazoti,  Odemir Martinez Bruno. 

This work describes a thinning algorithms study for automatic recognition ofprinted characters. A group of five thinning techniques were implemented: Hilditchalgorithm's, Stentiford algorithm's, Zhang-Suen algorithm's, Holt algorithm's andmultiscale thinning. A set of distinct printed characters was used in computationalexperiments. The resulted information of thinning process (the skeleton) was transformedin a feature vector through a remonstration technique. A neural network was trained toperform the character recognition. Every one developed thinning algorithms was evaluatedwith score classification (correct / incorrect printed characters recognition).

http://www.lbd.dcc.ufmg.br/colecoes/wvc/2005/0014.pdf

Processando consulta.