Panorama Stitching Technology Research

Virtual environment building based on image modeling and rendering is the hot point of the research on virtual reality technology. In the IBR technology, a 3-D scene is set up by using image-composing instead of 3-D geometry modeling, so it can speed up scene rendering and is highly efficient in constructing a specific virtual scene. Since panorama image is an image representation of virtual environment, panorama image mosaic technology has important applications in Virtual-Reality and Computer Vision and is the key of IBR.On the basis of the analyses of the existing algorithms, this thesis summarizes several main algorithms implemented in the research work. Firstly, we project images onto a cylinder and a spherical surface in order to achieve optic consistency. Secondly, we implement a feature-based stitching algorithm using robust RANSAC algorithm. Thirdly, to solve ghosting difference, the algorithm based on dynamic programming is used to find a best seam-line to stitch the images. In addition, a color blend approach based on tower structure can make the color transit smoothly. Lastly, a probabilistic model for image match verification is used. This gives us a principled framework for rejecting noise images and recognizing panorama in an unordered image set.After summing up the research work, this thesis points out some future research works.