Medical Image Registration Based On Wavelet Transform And Resampling

With the development of medical imaging equipment and technology, a large number of medical images such as CT imaging, MRI images, and so on, were put into clinical medicine. In clinical medicine, medical image registration technology is used to provide basis for clinical diagnosis and treatment. Therefore, the research about the medical image registration is very important for clinical medicine.The main work of this paper can be summarized as follows:A novel medical image registration method based on wavelet transform is proposed, in which the method of image dimension reduction combined with wavelet transform is adapt. The proposed method includes three processes, namely image dimension reduction, coarse registration and fine registration. The resolution of the registration images is reduced by the method of resampling. The purpose of the reduction of the images' resolution is to reduce the amount of calculation in the process of subsequent registration. In the process of coarse registration, edge contour features of images which were reduced resolution are extracted based on Canny operator and scanning line method. Furthermore, the image registration is completed by using the principal axis method, and the result of coarse registration is the basis of fine registration. In the process of fine registration, reference images and target images are decomposed by wavelet transform, then, the high frequency subband images(HL) are registered every other layer. At last, the reference image and the target image are selected as registration objects in the fine registration process. Therefore, the whole registration process is completed. The experimental results show that medical image registration method based on wavelet transform has higher registration precision. And also has high registration precision in incomplete image registration.In addition, according to the multi-scale method of wavelet transform, a medical image registration method based on multi-layer resampling is proposed. This method uses high frequency sampling operator to extract sampling image of each layer at first, then the sampled images are registered layer by layer. The experimental results show that the registration result error is within 0.1 pixel unit.