Warped Document Image Mosaicing Method Based on Inflection Point Detection and Registration

Abstract

With the popularity of digital camera and the application requirement of digitalized document images, using digital cameras to digitalize document images has become an irresistible trend. However, the warping of the document surface impacts on the quality of the OCR (Optical Character Recognition) recognition system seriously. This paper presents a warped document image mosaicing method based on inflection point detection and image registration. This method mosaics two warped images of the same document from different viewpoints. Firstly, one of the two document images is grizzled, binarized, dilated. Then text line extraction, characteristic sample point detection, curve fitting, and inflection point determining are performed. At last, image registration based on template matching, TRS transform, gray-level interpolation and image mosaicing are done for the two images. After mosaicing, the distortions are mostly removed and the OCR results are improved markedly. Experimental results show that the proposed method can resolve the issue of warped document image recognition more effectively.