Algoritmo sequencial para reconhecimento de numerais manuscritos desconectados utilizando redes neurais
AUTOR(ES)
Natanael Rodrigues Gomes
DATA DE PUBLICAÇÃO
1996
RESUMO
The main difficultyin handwritten character recognition consists in developing methods that provide a high recognition rate, although of the large degree of variability of the characteres. This work presents a system for recognition of disconnected handwritten numeraIs, based in analysis of the topology and distribution of pixels from the numerals, and application of a discrete Hopfield neural net used as associative memory. ln the system, the process of classification is divided in two stages. ln the first stage, the unknown numeral is classified considering features extracted from its topology and distribution of pixels. lf it is not possible, due to distortions and noise in numeral image, the classificationis effectuated in the second stage, via Hopfield netoThe Hopfield net is implemented of two manners. ln the first manner, the net has weights calculated by projection method and, in the second manner, the net has weights calculated by sinthesis procedures to linear systems operating in the saturated mode (LSSM systems). The system is tested with 1500 handwritten numerals. A recognition rate over 85% is obtained with the system making use in the second stage of the Hopfield net implemented by the first manner. A recognition rate over 84,4% is obtained with the system making use in the second stage of the Hopfield net implemented by the second manner
ASSUNTO(S)
reconhecimento de padrões redes neurais (computação)
ACESSO AO ARTIGO
http://libdigi.unicamp.br/document/?code=vtls000104130Documentos Relacionados
- HANDWRITTEN DIGITS RECOGNITION BY NEURAL NETWORKS
- Reconhecimento de padrões de falhas em motores trifásicos utilizando redes neurais
- Arquitetura de redes neurais para o reconhecimento facial baseado no neocognitron.
- Ambiente para otimização de redes multimídia utilizando algoritmo genético
- Reconhecimento de voz para comandos de direcionamento por meio de redes neurais