Geometric correction method for Tibetan woodcut document images

Abstract

In the information age, a large volume of Tibetan document images in woodcut edition have been produced. Geometric correction for Tibetan images is the basis of document analysis and character recognition, to which the traditional methods cannot be easily applied because of the complexity and variability of page deformation. To solve the above difficulties, we propose a new geometric correction method that is different from the existing methods. Firstly, we give a generic definition of piecewise strategy for image geometric correction by taking into full consideration the difficulty of seamless joint between sub-images. Then, we provide an effective geometric correction solution for Tibetan document images in woodcut edition, and theoretically prove the condition of seamless splicing between sub-images. According to this condition, appropriate reference points are selected to partition a Tibetan document image into several sub-images, then these sub-images are corrected by projective transformation, and finally the corrected sub-images are spliced to compose a corrected Tibetan document image. Experiments were conducted on real Tibetan woodcut document images by comparing the proposed method with classical algorithms including the FFT (Fast Fourier Transformation) method, the Hough method, the projection profile method and the cross correlation method. The visualized observation results and the statistical results proved that the proposed method can obtain a better correction performance, and took less than half the time of the comparison methods.