Research On SIFT-based Panoramic Image Mosaicing Algorithm

Panorama image mosaic is a process of combining multiple images with overlapping parts in the same scene.It has important applications in computer vision,remote sensing image information processing,aviation military,virtual reality.Different color image preprocessing,rapid matching of feature points,geometric correction and registration of different view angles,determination of suture lines,and image fusion under different lighting conditions in the process of panoramic splicing are all key techniques.This paper describes these key technologies from two aspects.Firstly,stitching of two images is focused on in this paper,where SIFT(Scale Invariant Feature Transformation)is used as the feature point and their fast matching strategy is studied.Because a large number of SIFT points can be produced in feature extraction process,this will cause the problem of long time consumption and redundant errors in the matching phase.This paper uses the BBF search algorithm instead of the traditional KD-tree algorithm to improve the matching efficiency,and then through the RANSAC algorithm to reduce the number of mismatch occurrences.In the experiment,simple-textured images and complex-textured images are selected for the splicing test.The optimization of the feature point matching process is demonstrated by the change of the number of SIFT feature points,and this result is compared with the original algorithm to verify the splicing accuracy and speed-up of the algorithm.Secondly,this dissertation studies the problem of panorama stitching of multiple chroma disequilibrium images under different lighting conditions in the same scene,and proposes a solution for image chroma and brightness equalization based on histogram specification;and it uses the statistical main direction of SIFT feature points as reference to calculate geometrical rotation angles between images;afterwards,the determination of sutures in overlapping areas and image fusion algorithms are described.In the experiment,three types of scenes,including mountain pass,traffic intersection,and building,are selected to test the algorithm and prove the effectiveness of the proposed algorithm.At last,the paper analyzes and summarizes the shortcomings of the algorithm,and puts forward the improvement measurements for the next step.