Research On Heterogeneous Image Registration Technology Based On Feature Points And Its Application

Image registration is the core of image fusion,and is also a hot research topic.It is widely used in the field of medicine and can be used for disease diagnosis and evaluation of a treatment plan.The registration accuracy is the primary evaluation index of medical image registration algorithms.However,due to the interference of medical imaging equipment’s own instruments and the complexity of the imaging environment,the resulting medical image has a lot of noise,which make it difficult to guarantee the accuracy in image registration.There are many problems to be solved in the registration speed and robustness of medical image registration algorithms.According to the research on registration algorithm,this thesis proposes a new feature-based heterogeneous image registration method.1.Introducing the basic principles of heterogeneous image registration and the commonly used algorithms and their advantages and disadvantages.Then focus on the Harris corner detection algorithm,analyze its existing problems,and improve the Harris corner detection algorithm based on the phase correlation algorithm.2.The improved Harris algorithm and SIFT operator are used to extract image feature points,the two-way matching method and RANSAC are used to remove errors,and the simulated annealing algorithm is used to obtain the transformation matrix.Finally,rigid registration is used to achieve image registration and fusion.3.MRI and CT brain images are used for simulation experiments,single-modality and multi-modality image registration experiments,and adaptive analysis based on point feature image registration is performed to obtain suitable reference images.Experiments prove that the improved Harris algorithm can detect corners more accurately and effectively,and the algorithm has higher speed and robustness.Heterogeneous image registration algorithm based on feature points proposed in this thesis has significantly improved image registration accuracy,algorithm speed and robustness compared with traditional registration algorithms.