Research On Feature-driven Image Registration And Image Mosaic Of CT Images

Medical Image Registration is one of hot research topics in the field of medical image. It has great significance in the clinical diagnosis and treatment. Medical image registration is for a medical image to find a (or a series) space transformation so that the corresponding points in this medical image will reach consistent with another medical image in the same place on both images. Many algorithms of medical image registration have been proposed. They are generally classified based on gray-driven method, modal-driven method and the hybrid algorithms.In the imaging process,the inspected area is often far greater than the image detector, the film must be divided into several parts, and then spliced together in accordance with certain rules, that is, medical image mosaic. Medical image mosaic has wide range of applications in the field of medical image. It can be solved the problem of unable to get the whole image because of the limited vision,and provide a better basis for the diagnosis. According to their methods, the medical image mosaic can be divided into the following categories: transform domain-based approach, the method based on image intensity and feature-based methods.As the medical imaging technology, computer science and technology continues to develop, the computer intelligent medical image diagnosis is one of the most important goals of medical image processing and analysis. To realize intelligent diagnosis or computer aided diagnosis, information comparability and information integrity are essential, of which the technology involved is image registration and image mosaic technology. Medical image registration and medical image mosaic are two hot and difficult research areas in the medical image field.Aim at the requirement of constructing a Computer Aided Diagnosis system for brain diseases based on High Resolution CT images, the research of this thesis is focused on the feature point-driven image registration and image mosaic technology.Based on the deep discussion and analysis for the methods of feature point automatic extraction, this paper put emphasis on the Scale Invariant Feature Transform (SIFT) algorithm. The matched points of two images were found, and paved the way for feature-driven image registration and image mosaic. These matched points were found by SIFT algorithm, achieving automatic matching feature points. For the image registration, we focused the research on the method of feature point-driven image registration. Based on the thin-plate spline interpolation algorithm, we realized the registration of two CT brain images, and also designed a globe feature vector based method to remove the error matched points of images that required for registration. Finally, primary research on medical images mosaic was conducted by using the method of feature point-driven image mosaic, and complete algorithm and programming procedures were proposed for two CT chest images mosaic. In accordance with the characteristics of medical images, the error matched points of images that required for mosaic were removed by random sample consensus (RANSAC) algorithm and statistics slope algorithm.