DEWARPING DOCUMENTS USING C2 CONTINUOUS BOUNDARY ESTIMATION

Hand-held mobile devices are commonly used to capture documents. Physical distortions in such captured document images make it difficult to extract information automatically. Uneven illumination and different lighting condition make the process more challenging. The problem is even more severe when we capture book pages. In this paper, we propose a novel document dewarping algorithm that works as follows. First, a document boundary (BD) mask is computed from an input image using a deep learning model which is further used to compute the document corners. Using the document corners, several equidistant points are estimated along each of the four document sides. These equidistant points are used to compute smooth outlines independently along four document sides using C2 continuous B´ezier curve. Finally, a mesh is computed using the B´ezier curves and the document is dewarped. The proposed algorithm is compared with the state-of-the-art algorithms in the literature and the results show the efficiency of the proposed algorithm. The low computational complexity of the algorithm makes it suitable to be used in any mobile devices.