Skip to Main Content
The performance (in terms of accuracy and speed) of present day document analysis systems, both handwritten and printed, depends on the preprocessing stage. In the existing literature we observe accuracy against speed trade-off. That is, superior binarization accuracy is arrived at the cost of increased processing time. The present paper proposes an improved binarization approach, which after a Pre Segmentation Binarization uses a state-of-the-art “fringe map” based text line segmentation algorithm to define text context. This text context is exploited to provide a more refined localized binarization (Post Segmentation Binarization). This unique combination gives benefits of both speed and accuracy with a scope of leveraging through parallel processing. It may additionally be used on skewed or warped camera captured document images. Experimental tests on ICFHR-DIBCO 2009, 2010 and 2011 databases of Printed/ Handwritten Document Images have shown promising results.