Localização do CEP no bloco do endereço em envelopes postais

AUTOR(ES)
DATA DE PUBLICAÇÃO

2005

RESUMO

This work presents a method for Zip Code location on address block of handwritten postal envelopes and also evaluates different strategies of modeling the components present in the address block. The proposed method starts by segmenting the address block into lines in order to execute the segmentation and recognition of the components of each line. Afterwards, the location of the ZIP CODE consists of searching for a sequence of 5 digits plus a sequence of 3 digits, separate or not by a dash. Hidden Markov Models and the Level Building Algorithm are used for finding the best segmentation-recognition hypothesis of the line being processed taking into account implicit segmentation-based strategy. In a second stage, the segmentation-recognition hypotheses are verified and re-classified. The address block presents a great variety of information (words, digits, character, lines, noises, etc.) hindering the modeling task. To deal with this problem, different modeling strategies for the address block components are evaluated. The best results were obtained by using the following Meta- Classes: M09 (digits from 0 to 9), MPC (words and characters), MH (dash), and MCEP (word CEP). The models constructed for the first stage of the system (segmentation-recognition) provide a recognition rate of 80.64%. Similar models trained with complementary information for the second stage of the system provide 85.38% of recognition rate.

ASSUNTO(S)

ciencia da computacao código postal processos de markov informática

ACESSO AO ARTIGO

Documentos Relacionados