Segmentation of Meaningful Text-Regions from Camera Captured Document Images

In the era of digitization, digitizing books, magazines, journals etc. are very much beneficial for humans. The scanning of a thick document and warped pages in the poor luminance cause long dark region alongside the margin. Imperfect thresholding during binarization also causes noise blocks. This paper proposes an approach for the segmentation of meaningful regions from camera captured document images based on projection profile and connected component analysis. The approach is tested on Tobacco-800 and Born-digital Warped Bangla Document Image Dataset (BdWBDID); The result is encouraging.