Research On Image Registration And Stitching Technology In Gobi Scenes

Image stitching is one of the most fundamental issues in computer vision and image processing,which emphasizes a process of aligning and combing a set of images into a high-resolution photo-mosaic.In order to stitch aerial images in gobi scenes effectively,this thesis aims to develop the novel image registration and stitching algorithms based on local feature descriptors.The main works and innovations of this thesis are as follows:1)This thesis analyzed the key technologies of image stitchingFirst,the theoretical analysis and concepts for image stitching are introduced.Then,several popular stitching algorithms are investigated.Finally,a description of the special technologies of image stitching in gobi scenes is presented.2)Region curvature histograms descriptors for image registrationIn order to represent the local features more effectively for image registration in gobi scenes,we propose a region curvature histograms based feature descriptor(R-CurvSift).The proposed method firstly introduces the maximally stable extremal region detector for invariant region detection,and then proposes a curvature histograms descriptor to increase the performance of image registration in gobi scenes.3)Image stitching based on local weighting projective transformAfter extracting the putative keypoint correspondences by R-CurvSift matching,a new local stitching method is developed.Specifically,in the proposed method,inliers are firstly extracted from candidates by guided sampling scheme,and the global homography is estimated with inliers.Secondly,a new local weighting projective transform model is presented to refine the homography estimation.The new transform model is a context-aware warping scheme which is more suitable for detail preserving in image stitching.