Automatic document image mosaicing algorithm with hand-held camera

This paper presents an image mosaicing method for camera-captured document images, and it can be used to stitch multiple overlapping document images into a large high resolution image. First, we use the nearest-neighbor(NN) clustering technique in document skew rectification to locate the horizontal vanishing point of the text plane. Secondly we partition the image into multiple overlapping blocks centered with the centroid of each connected component(CC), and propose a run-length opening algorithm(RLOA) to compute the local orientation of vertical character stroke(VSB), which is used to locate the document's vertical vanishing point. Thirdly, a three-step hierarchical rectification method is proposed to rectify document images. Finally, it uses local alignment constraints of all the overlapping image pairs to construct global alignment model, thus, to eliminate the error accumulation effectively. This method is unique in not calibrating the internal and external camera parameters in advance and not restricting the camera position, and it can produce a high resolution and accurate full page mosaic from small image patches of a document.