Research On Performance Enhancement Of Point Set Registration Algorithm Based On Structural Constraints

Point set registration is one important research topic in computer vision,pattern recognition,medical image analysis,etc.The goal of point set registration is to find the correspondence between a model point set and a target point set.In practical application,many unfavorable factors,such as deformation,rotation,outliers,noise,occlusion,and so on,may obviously reduce the accuracy and robustness of the point set registration algorithms.To address the above issues,two works have been carried out in this thesis.(1)An effective point set registration approach is proposed by combining local and global structure constraints.First,the composite weight coefficient is designed via the amplitude and projection of the vector from the reference point to its neighbor point.Then,a local structure constraint is constructed by using a linear combination of the vectors of neighbor points.A Gaussian mixture model is established by utilizing the local structure constraint and a global structure constraint based on the motion coherence.Finally,an iteration coefficient update strategy and an expectation maximization algorithm are combined to solve the parameter optimization problem of the proposed model.Compared to the existed point set registration approaches,the proposed model is more robust due to the effective constraints.The experimental results on some widely used data sets demonstrates the effectiveness of the proposed model.(2)A robust point set registration algorithm is proposed based on the decomposition of spatial transformation and corresponding structure constraints.First,the spatial transformation is decomposed as two types of transformations,i.e.an affine transformation and a non-affine transformation.In order to enforce the smoothness of the spatial transformation,the square norm of the kernel Hilbert space is adopted as a coherent constraint for the non-affine transformation.In addition,for the affine transformation,a constraint term of the parameter vector is applied to improve the robustness and efficiency.Furthermore,the graph Laplacian is used as the manifold constraint to capture the spatial geometric information of the point sets.The experimental results verify that the proposed point set registration model is more robust to deformation,rotation and outliers.