Segmentation of Printed Gray Scale Dot Matrix Characters

Conference paper


Martin Grafmüller
Jürgen Beyerer


Jorge Baralt, Nagib Callaos, Shigehiro Hashimoto, William Lesso, Dale Zinn (eds.), Proceedings of the 14th World Multi-Conference on Systemics, Cybernetics and Informatics (WMSCI), vol. II, 2010.




14th World Multi-Conference on Systemics, Cybernetics and Informatics (WMSCI), Orlando, USA, June 29 - July 2, 2010

There are many kinds of printed dot matrix fonts. They mainly differ in characters' structure since some of them are built with touching dots and some not. Two more variations an efficient character segmentation method has to cope with are fonts printed in italic and illumination changes depending on the environmental conditions. Hence, a procedure is introduced that uses the projection profile of the mean of several least gray values to estimate the skew angle of text lines and the characters' slant angle. Furthermore, based on the projection profile the decision boundaries for line and character segmentation are estimated with a least squares estimator. This has the advantage that the boundaries are adapted for every image. Finally, skew and slant angles are corrected and the characters are segmented.

The robustness and accuracy is shown in experiments on data captured with an industrial camera.