Character feature extraction using polygonal projection sweep (contour detection)

Abstract

It is presented in this paper a new approach to the problem of feature extraction. The approach is based on the edge detection, where a set of feature vectors is taken from the source image. The images under this investigation are considered to be manuscript characters and the features are obtained by the distance from the contour of each character to several observation points placed around the image. Such observation points are arranged along different geometric polygons built in a way to surround the image. The approach is evaluated against the naïve bitmap matrix considering different types of polygon. The discrimination power of each method is computed using both statistic and neural network entries. The proposed approach provides also good response to the scale, rotation and translation problems in addition to discrimination. © Springer-Verlag Berlin Heidelberg 2001.