Studies On Image Mosaic And Its Application In Virtual Reality

Panorama is an image with a wide-angle view.It can be taken with special cameras;it can also be constructed through image alignment techniques from regular hand-held cameras.The technique to create panoramas is among the hottest topics in field of IBR(Image Based Rendering). Full-view panoramic image mosaics has become a important way to study virtual reality and computer vision.The thesis focuses on study of Studies on Image Mosaic and its Application in Virtual Reality, The steps of panoramic image stitching are emphasized.Virtual reality and its scene rendering methods,image acquisition, image matching,image fusion are studied emphatically. The main works are detailed as followed:(1) The research status of virtual reality and image mosaic are reviewed.The existing problems of panoramic image generation are analyzed.The flow chart of panoramic image generation is shown.(2) The meaning,characteristic,catalog of virtual reality are dissertated.Virtual reality scene rendering methods are analyzed and compared.The thesis uses image-based rendering method that is explained in detail.Image-based rendering method and the characteristic of image-based rendering panorama video are introduced.(3) This capter introduces the key technologies of image mosaic:image model, transforma- tion model,image matching,image fusion and so on.(4) The current image registration methods are introduced and their advantages and disadv- antages are analyzed.The SIFT algorithm is improved by researching on sift feature and belief propagation algorithm. First,bacause of lighting and glisten,different of image areas have different threshold.So the adaptive threshold algorithm is presented in this thesis. While SIFT descriptors accurately extract invariant image characteristics around keypoints,it completely ignores the geometric information among descriptors. In this paper, we formulate keypoint matching as a global optimization problem and provide a image matching algorithm based on sift feature using belief propagation.Experimental results show significant improvement overprevious approaches.Compared with the original algorithm,the new algorit- hm can match images accurately and quickly.(5) A bundle-adjustment algorithm is applied to prevent image distortion when there are more than two source images registrating,disregarding the accumulated errors caused by concatenation of pairwise homographies.The thesis makes an improvement on overlapped area linear transition method,which combines with light intensity compensation method and uses cosine form to adjust transition factor,solving the effect on obvious changes in light intensity,and have made satisfactory effect. (6) Finally,the virtual reality system based on image-based rendering method is introd- uced,mainly researching on its structure and the steps of coming ture.