Low Quality Mobile Image Data Processing Under Uneven Shading - Separating and Cleaning Text Lines and Graphic Regions in mobile Color Document Image

This paper proposes a simple approach for extracting texts from graphic regions in low quality color document images taken by smart phones or other mobile devices with cameras. An algorithm first computes an edge map by the Canny edge detector. All textual and non-textual regions are then analyzed heuristically based on their connected components(CC). A 2D histogram is calculated to estimate the frequent width and height of connected components. After grouping the CCs according to association rules, the CCs in which the width or height levels are then measured as extremely large or small are assigned as non-textual regions. The remaining CCs are then extracted as text regions. The results of our experimentations demonstrate that the proposed approach performs with plausible consistency.