Font Size: a A A

Research On Matching Algorithm For Image Fast Stitching Using Feature Points

Posted on:2012-05-02Degree:MasterType:Thesis
Country:ChinaCandidate:S W HanFull Text:PDF
GTID:2218330362452934Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
The technique of image stitching generates one picture with many images that have multipleoverlapped parts, in order to expand the scope of the viewport. In computer vision, videosurveillance, remote sensing and other research areas, the key problem is how to obtain a largefield of view scene photos without declining the quality of the photos. Furthermore, when usingordinary camera to shoot wide view of the scene image, we must obtain the full scene byadjusting the camera focus, but that will decline the quality of photos. The image stitchingtechnique is an effective method to solve this problem. Matching Algorithm for Fast ImageStitching Using Feature Points, which is main research content of this paper, has a goodpracticality and application prospect in the field of computer vision.On the base of image preprocessing, Harris method and the matching method which isbased on gray correlation matching, In this work, the use of Euclidean Distance of the ClusterPre-screening methods is mainly based on the principle that euclidean distances betweenmatching points are the same or similar. By using a simple cluster filter out the feature points'union which contains the most matching feature points from all the filter unions of the featurepoints, the pairs of feature points which are in the feature points'union can be considered asmatching points. On the contrary, the others are not the matching points, hence removing theothers from the candidate matching points. RANSAC algorithm is used in this paper to matchpoint set for further precise matching. In order to achieve seamless image stitching, the otherproposed algorithm which combines the LM weighted fusion algorithm with the Laplacianpyramid image fusion algorithm is used for image fusion.The paper mainly analyses the image matching and image fusion in-depth study base ontheoretical analysis and the experiments. Experiments show that the algorithm effectivelyreduces the number of iterations and improves the execution time of RANSAC algorithm by85%, without reducing the matching points. Thus, this algorithm greatly improves the efficiencyof matching algorithm on the premise of promising the matching accuracy. In addition, theproposed algorithm combines the LM weighted fusion algorithm with the Laplacian pyramid fusion algorithm, and effectively eliminates the ghosting and the abrupt change of brightness.Finally, the proposed algorithm achieves seamless image stitching.
Keywords/Search Tags:mosaic, clustering pre-screening, rannsac, lm weighted fusion algorithm, laplacian pyramid
PDF Full Text Request
Related items