Restoration of Images Scanned from Thick Bound Documents

While scanning thick bound documents, the pages are not flat on the document glass of the scanner. The physical deformation of the scanned page can results in two kinds of degradation for the scanned image. One is the shadow incurred near the spine of the book; and another is the text being bended. In this paper, we propose a method to combine the information both from the scanned image and from the geometric distortion to remove the shadow as well as restore the warped words to the right positions. First, the shadow is removed by patch-based auto-threshold binarization. Then the central lines of text are directly extracted from the binarization image. This goal is achieved by using vertical projection function, valid bounding boxes, and markers. Finally, the bended lines and the warped words are restored by the geometric parameters evaluated from the central lines and the piecewise quadrilateral map. Experiments show that the proposed algorithm gives satisfactory results.