Segmentation of Hardwritten Zip Code Using an Artificial Neural Network

概要

論文の詳細を見る
In this paper, we propose a useful method for segmenting the character area of hand-written zip codes using an artificial neural network. This method, which is constructed by three networks corresponding to each digit, can segment each character of three-digit zip codes without specific processes for faded characters, continued characters, and so on. The internal structure of the network is investigated in detail. As a result of the investigation, it is confirmed that the network detects the character area not only from the feature of the corresponding digit but also from that of the other digits. The ability of this method with second search is evaluated as a segmentation rate over 99% for 979 test images.
社団法人電子情報通信学会の論文
2002-03-01