Research And Implementation Of Medical Image Registration Method Based On Improved Canny And Optical Flow Model

Medical image registration is the seeking of a proper space transformation to achieve the spatial matching and registration of two images, finally attaining image fusion. Recently, medical image registration is becoming more important in the area of medical image processing, and non-rigid registration is a hot research topic among medical image registration.This paper does a deep research on non-rigid medical image registration algorithm, describes the development and research status of the medical image registration technology, analyses the theory and methods of the non-rigid medical registration systematically, and discusses the concept, category, stage, method and assessing methods of the medical image registration. On that basis, an algorithm based on improved Canny and Demons optical flow model for non-rigid medical image registration is proposed.This paper introduces the fundamental theory of the traditional operator and Canny operater edge detection. On this foundation, an improved edge detection based on the adaptive Canny operator is proposed. It adaptively changes the coefficients of the filter according to the revulsion characteristic of image pixels gray scale, and the edge can be sharpened in the smoothing process. The results of the experiments improve the performance in extracting image edge details and suppressing false edge.The paper focuses on the research of optical flow model and the original Demons optical flow algorithm, proposes the improved Demonos optical flow algorithm which not only improves the optical flow constraint equations but also enlists the reference image gradient as the new internal force to drive the image approaching toward the regerence image. This method can solve the blurring problem during registration of the original optical flow model, and adds the ability of handling image with large deformation.The algorithm in the paper combines the coarse registration with the fine registration. The coarse registration is affine transformation based on the improved Canny operator, which can eliminate the diference of the overall rotation and translation between the reference images and image to be registrated. And the fine registration is optical flow registration model based on the improved Demons, and improves the registration accuracy.The development tool in the paper is VC++based on the OpenCV. This paper is based on many experimental data and comparative registration results between the images of MRI and MRI, CT and CT, PET and PET, CT and MRI, PET and MRI, PET and CT, together with the experiments of feature extraction, coarse registration and fine registration, shows that the improved method in this paper can solve the blurring problem and handle image with large deformation. As a result, the superiority of the algorithm in the paper is proved.