For the gray-level image of government resource document, a linear transform was employed to enhance the image contrast. A spatial filter was applied to eliminate image noise. After this preprocessing, the threshold surface T1 was computed by Bernsen algorithm, and the global threshold T2 was calculated by modified Otsu method. On the basis of T1 and T2, other three thresholds were defined, which include the broken stroke value T3, the average value T4 of neighborhood and the union value T5 between global and local. Then the gray-level image was binarized through the combination of these five values. Our experiments showed that the proposed method using these five thresholds was adaptive to various government documents. By ghost artifacts eliminating and the broken strokes mending, it's benefit to OCR.
OCR Oriented Binarization Method of Document Image
2008 Congress on Image and Signal Processing ; 4 ; 622-625
2008-05-01
1635764 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Document image cleanup and binarization
SPIE | 1998
|A Noise Attribute Thresholding Method for Document Image Binarization
British Library Conference Proceedings | 1995
|A system for document binarization
IEEE | 2003
|A System for Document Binarization
British Library Conference Proceedings | 2003
|Document image binarization based on texture analysis [2181-25]
British Library Conference Proceedings | 1994
|