Video Mosaicing for Curved Surface by 3-d Reconstruction Using Feature Points

Video mosaicing methods have been widely used for capturing high resolution document images with low resolution video cameras [1, 2]. In these methods, given a video where partial images of the target document are captured in each frame, pairwise registrations between two successive images are estimated, then, by warping all the images to a reference frame (in general, the first frame), a high resolution document image called a mosaic image is synthesized. These methods, however, assume that the target is a plane, thus will not work for documents with curved surface. An example of a document with a curved surface is shown in Figure 1. We propose a novel video mosaicing method for curved surface based on 3-D reconstruction. With the proposed method, the plane mosaic image of the target document is generated, even if the document has a curved surface. We have developed a demonstration system based on this method. In the following sections, we briefly describe the proposed method, and then introduce our demonstration system.