On the Recovery of Oriented Documents from Single Images
Paul F. Clark, Majid Mirmehdi · Bristol Research (University of Bristol) · 2002
A method is presented for the fronto-parallel recovery of text documents in images of real scenes. Initially an extension of the standard 2D projection profile, commonly used in document recognition, is introduced to locate the horizontal vanishing point of the text plane. This allows us to segment the lines of text, which are then analysed to reveal the style of justification of the paragraphs. The change in line spacings exhibited due to perspective is then used to recover the vertical vanishing point of the document. We do not assume any knowledge of the focal length of the camera. Finally, a fronto-parallel view is recovered, suitable for OCR or other high-level recognition. We provide results demonstrating the algorithm's performance on documents over a wide range of orientations.