Reconhecimento de caracteres numéricos manuscritos com autômatos celulares, através da noção de protótipos espectrais

Imagem de Miniatura
Data de publicação
Citações (Scopus)
Oliveira Junior, Carlos Candido de
Oliveira, Pedro Paulo Balbi de
Título da Revista
ISSN da Revista
Título de Volume
Membros da banca
Marengoni, Maurício
Macau, Elbert Einstein Nehrer
Engenharia Elétrica
The computational ability of cellular automata (CAs) has been investigated in many research initiatives in the literature. Among them, one possibility is their use as spatial pattern recognizers. Here, we attempt to use CAs to recognize handwritten numerical characters, by looking for CA rules which, running on different initial configurations that represent the same handwritten digit, would lead the temporal evolution to equivalent final configurations (defining the same spectral fixed point. The work uses a metric that allows the comparison between images based on their spectral similarity, drawn from a previous work, as well as a proposal therein on the way elementary CA rules are applied (in a crossed fashion) so as to allow the reduction in the size of the rule space involved. Conceptual structures are defined to support the recognition experiments, namely, the notions of spectral fixed point, spectral prototype, and a measure of a rule potential to perform well in the recognition task. The latter entailed from various studies on the attributes a good rule should account for, and, once defined, made it possible an evolutionary search in that space. Here we show some rules that yielded positive results in the recognition of the image used for training (i.e., those used in the definition of the recognition prototypes), and further results on the attempt to carry out real recognition experiments, with images not used for training, whose results are not yet satisfactory.
autômatos celulares , algoritmos genéticos , protótipo espectral , automata cellular , genetic algorithms , spectral prototype
Assuntos Scopus
OLIVEIRA JUNIOR, Carlos Candido de. Reconhecimento de caracteres numéricos manuscritos com autômatos celulares, através da noção de protótipos espectrais. 2006. 144 f. Dissertação (Mestrado em Engenharia Elétrica) - Universidade Presbiteriana Mackenzie, São Paulo, 2006.