Research Of Multimodal Medical Images Registration Based On SURF And Optical Flow

With the development of biomedical engineering and computer technology,a variety of multimodality images can be obtained by the imaging equipment,such as CT and MRI.It is good for clinical diagnosis and treatment.Due to the differences of multimodality images,we often combine multimodality images to improve the clinical medical diagnosis and treatment.The image registration refers to finding geometrical transformations between two images so that the corresponding points of two images are spatially "consistent".The image registration is an important research in medical image processing,and it is the basis of image fusion and three-dimensional reconstruction.A new non-rigid registration algorithm based on optical flow and SURF in this paper and hardware acceleration technology for image registration algorithm based on CUDA architecture.The main work of this paper as following:1.The research background and status of the medical image registration are introduced in this chapter.The main problems challenges of the multimodality image registration based on optical flow field are pointed out.Then,the basic theoretical knowledge of image registration is introduced.Last,the preprocessing,interpolation and feature space and the framework of image registration are introduced in detail.2.Because the image registration based on optical flow requires that consistent grayscale of the floating image and the reference image,so it is not suitable for the multimodality images.Therefore,the histogram specification is used to the image processing.After the histogram specification,the similar modal of images is beneficial to the following image registration.3.Due to the traditional optical flow model has the requirement of “gray consistency”,the registration algorithm based on the traditional optical flow requires less difference in structure and gradation between the images to be registered.However,there are large differences between medical images of different modalities,so image registration algorithms based on traditional optical flow are not suitable for multimodal medical images.In order to solve the problem based on traditional optical flow model,this paper uses the hierarchical registration method to match the SURF algorithm and medical image registration algorithm based on traditional optical flow.Firstly,the SURF algorithm is used to extract the features of two medical images,and the feature points are matched to complete the initial registration between medical images to reduce the difference between images.Then,the medical image registration algorithm based on optical flow field model is used to accurately achieve the matching of the images obtained by SURF algorithm.First,the optical flow field between the input initial matching image and the reference image is calculated according to the Horn-Schunk model.The deformed field of the image is determined by the calculated optical flow field,and then the floating image is interpolated using bilinear interpolation to obtain final registered image.4.In order to improve the computational efficiency of the multimodality image registration based on the optical flow,the high-performance computing technology based CUDA was proposed in this paper,which makes full use of the GPU parallel computing under the CUDA architecture.And on the premise of ensuring the accuracy of the registration results,it can realize quickly the multimodality image registration based on optical flow.