Segmentation of text/image documents using texture approaches

Trygve Randen and John Håkon Husøy
Høgskolen i Stavanger
P.O. Box 2557 Ullandhaug, N-4004 Stavanger, Norway

To appear in Working Papers from Høgskolen i Stavanger, 1997
Also submitted to IEEE Transactions on Image Processing

Full paper


A fast and flexible method for segmenting text and graphics of document images is presented, assuming that text-regions have different texture properties than non-text-regions. As features we use measures of the local sub-band energies in the output of a computationally efficient critically sampled filter bank. First we develop an un-supervised approach, and we get good results, but even better segmentation is desired. We therefore extend our approach with a novel supervised segmentation scheme. The experimental results from the supervised method show the expected improvement over the results from the un-supervised method.