Título: Uma Solução Backpropagation Invariante a Escala, Rotação e Translação para o Reconhecimento de Caracteres
Autores: Varella, Luiz Eduardo Seabra; Passos, Emmanuel Piseces Lopes; Santos, Márcio Azevedo; Araújo, Ricardo Lomba de
Resumo: Character recognition represents one of the steps of major complexity in systems designed for the recognition of graphic documents (maps). The method suggested in this paper uses neural networks to recognize characters with variations of scale, rotation and translation. The method primarily consists of the extraction of invariant moments of image. These moments are subsequently submitted to a system of Backpropagation (multi-layer Perceptron and Backpropagation learning) networks. Such system consists of 466 networks, which classify the image characters. Results thus obtained reveal a high degree of accuracy to the extent that 97% of trained patterns were duly recognized.
Código DOI: 10.21528/CBRN1994-029
Artigo em PDF: CBRN1994-paper29.pdf
Arquivo BibTex: CBRN1994-paper29.bib