Maryam Golpardaz - Department of Electrical EngineeringKerman UniversityKerman, Iran
Hossein Nezamabadi-Pour - Department of Electrical EngineeringKerman UniversityKerman, Iran

Most of the existing algorithms in rectification ofgeometric distortions in document images are for Latin texts. Thesealgorithms are time consuming and show error for the skew anglesgreater than 40 degrees. Also they can't remove both skew andperspective simultaneously. In this paper, we propose an algorithmthat rectifies perspective and skew in Farsi document images in awide range of angles with an acceptable accuracy. The proposedalgorithm is applied in gray-scale images. First, we use a localthresholding method to extract the binary image adaptively fromthe degraded gray-scale document image. Then, skew angle of theimage is detected and it is rectified by rotation. Finally, we correctperspective in skewed document image with perspectivetransformation. The proposed algorithm can be applied in Latinand Farsi document images with any skew angles and perspectivedistortions. The experimental results confirm the effectiveness ofthe proposed method.

Perspective Rectification, Document Skew, TiltAngle, Connected Component Labeling