Dewarping of Single-Folded Camera Captured Bangla Document Images

Abstract

The document images captured through mobile camera get warped due to the curved surfaces like an open thick book page, notices posted on a cylindrical lamppost or a paper hanging from a notice board attached at top-middle. The very recent state of the art techniques for dewarping, work mostly for document images containing alphabetic script like English; these cannot handle all the variety of warping. This paper proposed a method for dewarping of single folded warped document images containing alpha-syllabary script like Bangla. Here the information of top line and baseline increase the robustness of the algorithm. Unlike other related methods, this work has not used any additional information like focal length, distance from the camera or view angle. Experimental evaluation with the collected dataset shows promising results.