H. Nezamabadi-Pour, E. Kabir and R. Azmi,
Volume 23, Issue 1 (7-2004)
Abstract
In this paper, a modified segmentation algorithm for printed Farsi words is presented. This algorithm is based on a previous work by Azmi that uses the conditional labeling of the upper contour to find the segmentation points. The main objective is to improve the segmentation results for low quality prints. To achieve this, various modifications on local baseline detection, contour labeling and segmentation rules have been applied. In an experiment, the correct segmentation rate was 97%. Based on the results obtained, a detailed error analysis is presented which should be useful for furthur research on this topic.
