Uma abordagem fractal para localização do bloco de endereço em segmentação de envelopes postais

AUTOR(ES)
DATA DE PUBLICAÇÃO

2005

RESUMO

With the increase of the computation equipment performance and efficiency, the challenges of automatizing and modernizing the Postal Automation are becoming each time more realistic. The computational tools inherited of the Area of Processing and Analysis of Images allow to acquire, to store and to execute applications with high degree of complexity. An approach based on the fractal dimension to automatically segment the block-address in postal envelopes is described. This approach aims to automatically separate and differentiate the block-address, and stamps and postmarks in postal envelopes. Initially, the fractal dimension computed in pixel-centered windows is attributed to each pixel of the image. After that, the K-means algorithm divides the fractal dimension image in three clusters: the background one, the noise one and the block-address, stamp, postmark one. The identification of each cluster is based on co-occurrence matrix. The quantitative evaluation of the segmentation is carried through comparing the result obtained for each envelope with the ideal result (strategy ground truth). This approach was tested on two postal envelope image databases - one Real and other synthesized - that aggregate some types of deep and layouts. The obtained results have shown that this approach is promising with an accuracy rate of 96% for the address-block with less than 7% of noise. Conclusions and future research are presented at the end of this work.

ASSUNTO(S)

ciencia da computacao serviço postal - automação processamento de imagens

ACESSO AO ARTIGO

Documentos Relacionados