Font Size: a A A

Research Of Local Feature Points Based On Image Registration Method

Posted on:2017-05-10Degree:MasterType:Thesis
Country:ChinaCandidate:C HuFull Text:PDF
GTID:2308330503960357Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
Image mosaic and fusion is one of the hot point research of image processing,pattern recognition and computer vision. Experts and scholars has been carried out extensive research on 3D reconstruction from image sequence, but mostly concentrated in the two-dimensional plane. When the shape, size, position and orientation of the target sequence of images is changed, It would be difficult to achieve when the original fully static two-dimensional image of the same size to match the stitching. For these problems, this paper focuses on the image local feature points registration method,mainly to complete the following work:1.Image stitching and fusion consult relevant literate,described and analyzed the main problems of image fusion research status and existence..2.Local feature points on image registration method of the key technologies,including image stitching technology overview, image stitching technology foundations of geometry, image transformation model, elements and methods of image matching,image stitching process.3. In view of the SIFT algorithm for multi-point and match mismatch slow problem,puts forward a kind of SIFT pixel dimension reduction bidirectional matching filter pretreatment method, before constructing DOG space SIFT pixels screening pretreatment, reduce useless feature points; When the feature descriptor is generated,the dimension reduction processing is carried out, the last registration of the use of restraint method to achieve two-way matching SIFT. The experimental results show that compared with the original algorithm the improved method can improve registration accuracy and matching speed.4. In view of the traditional algorithms of the ORB is uneven distribution of image,matching problem of high error, presents a combined with regional block and the ORB algorithm of image registration method. Before the ORB to extract the image feature points, search time by region partition method is used to reduce feature points; The improved RANSAC method is used to eliminate possible match. The experimental results show that the proposed method to ensure the stitching accuracy and improves the matching speed.
Keywords/Search Tags:image registration, pixel selection, dimension reduction, ORB algorithm, regional block, constraint registration
PDF Full Text Request
Related items