Study Of Panoramic Stitching

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)Panorama recognizing, also called image stitching or image mosaic, has been studied for a long time. Because of its variety and complexity, it has been studied widely. According to statistics of ISI (Institute of Scientific Information), there are over 1000 articles on such field.In this article, we introduce several image stitching algorithms both national and abroad and clarify them; we also describe the new technology and introduce some related knowledge about it. With previous research, we build an automatic spherical panorama recognizing system. The images to be stitched are captured with hand-hold cameras. After inputting several images, it can partition them into several parts, images in each part being overlapped, and stitch each part into a spherical panorama. According to such requirements, we design a panorama recognizing algorithm , fully considering robustness efficiency quality and so on. At first, SIFT (Scale Invariant Feature Transform) are extracted from each images and matched with KD tree; Secondly, unmatched feature pairs (outliers) are removed by the improved method based on RANSAC. Whether two images are matched or no is depended on the linear relationship between number of total feature pairs and number of matched feature pairs (inliers); Thirdly, Euler angle and other parameters are computed with LM method and images can mapped into a virtual sphere with these parameters; At last, remapped images are blended into a spherical panorama with multi-band blending method. We also build two application systems based on spherical panoramas, one is to view spherical panorama and the other is a scene tour system. They can be used for show housing, digital map and so on.