Research On Data Registration Algorithms Based On Spatial Structure Correlation

With the advancement of technology,data registration techniques are being increasingly applied in various fields such as image stitching,image enhancement,and 3D reconstruction.In practical applications,there may exist complex geometric transformations between the data,including rotation,translation,scaling,and nonlinear deformations.Additionally,when there are changes in the viewpoint of the object or scene or when partial occlusions occur,the extracted feature points often contain influencing factors such as noise and outliers.Consequently,image distortion or blurriness poses challenges in accurately matching the feature points and accurately estimating the transformation model.Point registration is a competitive data registration technique as it aims to capture the overall structure between two point sets purposefully.Typically,the point registration problem can be divided into two sub-problems: searching for correspondences between point sets and estimating the spatial transformation matrix.Searching the optimal correspondence is a classic combinatorial explosion problem,while estimating the transformation matrix is a continuous space optimization problem.Furthermore,the image feature matching process is susceptible to mismatches due to occlusions,deformations,and variations in lighting conditions.Despite the satisfactory performance of existing point registration and feature matching algorithms,there are still challenges in dealing with inaccuracies in matching,robustness issues concerning noise,outliers,and missing points.This thesis focuses on studying point registration methods based on the data space structure and techniques for removing mismatched correspondences.The primary approach involves utilizing global and local structural features between point sets for registration and eliminating mismatches.Several point registration algorithms and mismatch removal methods are proposed to overcome the limitations of existing techniques.The specific research work includes the following aspects:1.There exists a positional error between two feature point sets detected from two lowquality images,often accompanied by bilateral outliers that cannot form corresponding relationships.To address the aforementioned issues,we propose a novel point registration algorithm based on total least squares(TLS)with a dual term,which can correct errors-invariables and suppress multiple outliers.Such TLS-based criterion with single roworthonormal space transformation matrix includes two interesting dual(symmetrical)terms that can be conveniently exploited to suppress bilateral outliers.The proposed method simultaneously estimates bilateral transformations using the same row-standard orthogonal matrix and utilizes the dual term to suppress bilateral outliers,thereby enhancing the algorithm’s robustness against outliers.Experimental results demonstrate that the proposed algorithm outperforms existing methods in several computer vision tasks.2.The goal of non-rigid point set registration is to estimate the optimal correspondence between points,and then recover the non-rigid deformation between point sets in a specific way,typically by using a set of complex interpolation functions.Many existing non-rigid matching algorithms only utilize the local structure between point sets to a limited extent.To improve the accuracy of point set registration,a non-rigid registration method is proposed that leverages both the global structure and stable local structure of non-rigid shapes to constrain the registration.Specifically,we consider the point set registration problem as a probability assignment problem,with the probability determined by the Gaussian mixture model and the local structure of the point set.In particular,the Hausdorff distance is effectively used to measure the similarity of the local structures of point sets and adaptively assigns weights to the Gaussian models.Extensive experiments demonstrate that the proposed technique has higher registration accuracy than several other state-of-the-art algorithms when dealing with non-rigid registration problems,especially when the point set contains outliers and severely missing points.3.The goal of feature matching based on data space structure similarity is to establish accurate correspondences between feature points in different images depicting the same scene.To address the polymorphism of local structures,a false-match removal method using bilateral local-global structural consistency is proposed.This method incorporates the problem of mismatch removal into the framework of graph matching,constructs a global affinity matrix using local structural similarity and global affine transformation consistency,and optimizes it using a constrained integer quadratic programming method.Specifically,the weights of edges are constructed based on the signature quadratic form distance(SQFD)of the local structure,while the matching correctness of nodes and edges between the two graphs is described using local vector similarity.Furthermore,the consistency of the global affine transformation is evaluated by assessing the consistency of the local neighborhood affine transformation between different corresponding point pairs.Experimental results demonstrate that the proposed algorithm outperforms existing state-of-the-art methods in terms of accuracy and effectiveness.4.Image feature matching refers to the process of establishing correspondences between feature points in two or multiple images taken from the same scene using specific algorithms.Establishing accurate correspondences between feature points is essential for reliable matching point pairs,which in turn improves the accuracy of image registration.Typically,feature point matching relies on the consistent local structural information of point sets to estimate correspondences.To preserve reliable correspondences,a novel image feature matching method based on local structure consistency is proposed.The method introduces the Hausdorff distance to measure the similarity of local structures and combines it with the feature vectors of neighborhood structures to construct a mathematical model,deriving a closed-form solution with linear time complexity.In addition,a multi-scale neighborhood strategy is employed to retain reliable correspondences for image feature matching.Simulation experiments demonstrate the effectiveness of the proposed method in various remote sensing image feature matching tasks.