| Image mosaic technology is to generate a panoramic image with greater view and high-resolution from multiple images which have overlapping regions. And the technology widely used in remote sensing image processing, medical image processing, panorama generation, virtual reality and other fields is the research hot point of computer vision and image processing.This thesis discusses the stitching of multiple images taken by ordinary camera or other hand-by devices like mobile phone. Based on the research result of local invariant feature of image and panorama generation technology, an automatic image mosaic algorithm based on multi-local features matching is proposed in this paper.The algorithm detect the SIFT features and MSER features which are described by SIFT descriptor in the scale-space of input images; we use the BBF algorithm based on k-d tree to find the nearest neighbor and sub-neighbors of SIFT features and MSER features whose ratio is used to determine the matching relationships; based on RANSAC method to get transform relations excluding outliner, we deploy the probability theory method to determine the image is matched with other images which are selected from top images which have the most matching points respectively; we get several image subset having matching relationships from images, and processes the subset with bundle adjustment and photometric differences adjustment, finally we get a panorama from each subset through multi-band method.We finally test the algorithm by programming and the result shows that the algorithm is effective.The improvements of our approach:The algorithm detects MSER features in Gaussian scale-space to enhance its adaptability to scale changes and realizes the detection of image's local features in more robust way by combining the SIFT features. A simple but effective photometric differences adjustment algorithm is further proposed to generate more natural panoramic image of a scene with the images taken from different view points. |
PDF Full Text Request