An Improved Adaptive Document Image Binarization Method.pdf

Binarization MethodShuang-fei ZHOU Chun-ping LIU Zhi-ming CUI Sheng-rong GONGSchool of Computer Science and TechnologySoochow UniversitySuzhou, Jiangsu 215006, ChinaAbstract-Document image binarization is the basis of OpticalCharacter Recognition (OCR). For B.Gatos’ adaptivebinarization method exists the shortcomings, we propose animproved adaptive document image binarization method whichconsists of four steps. The first step is dedicated to a denoisingprocedure using a low-pass Wiener filter based on local statistics.In the second step, we use a first rough estimation of foregroundregions using binarization method based on the Laplacian-Gaussalgorithm. As a third step, we compute the background surfaceof the image by interpolating neighboring background intensitiesinto the foreground areas that result from the previous step. Inthe fourth step, we proceed to the final binarization bycombining information from the calculated background surfaceand the original image, and accomplish final binary. The methodhas good robustness for uneven illumination, using the algorithmextract foreground regions, it will get fewer lost strokes and canbe effective to retain the edge information. The experimentalresults show that the improved method has better characteristicsthan other four kinds of typical document image binarizationmethods.Keywords-Local adaptive binarization; Laplacian-Gauss; Uneven illumination document imagesI. IntroductionIn most document image analysis and understandingsystems, document image binarization is first carried out apre-processing step; its effects will seriously affect the qualityof the character segmentation and recognition. According tothe scope of the different computing, document imagebinarization method is divided into global and local method;global method determines a threshold based on document