Non-Rigid Medical Image Registration Method Based On Features And Gray

Image registration is the basis of image fusion, analysis, reconstruction, and especially in the field of medical image processing, where, image registration plays an important role. Effective and accurate registration method can help doctors do medical diagnosis, make operation plan, track of pathological changes, and evaluate the effect of the treatment.In recent decades, most of non-rigid registration methods are single methods, for example,based on the gray or based on geometric features. The related research materials and clinical applications show that two kinds of registration methods have some advantages as well as limitations. Gray-based registration method directly uses gray information, and is more suitable for monomodal image registration. However, for some complex tissue image, such as the heart or the liver, this method can not get good effects. While feature-based registration method can make full use of the geometric features, and is mainly suitable for multimodal image registration. But precision of the registration result depends on accuracy of the feature extraction.In this paper, we analyze the viscous fluid registration method gray-based deeply, In order to utilize feature information of the image in scale space to improve the precision of the viscous fluid registration method gray-based, we make use of characteristics of the scale invariant features and the structure context features, to make up the errors for the complicated structure of the image using the viscous fluid model registration method, in this thesis, we design a hybrid algorithm combining the gray information and geometric features of the image.This paper expands the scale invariant features and the structure context features from 2-dimension to 3-dimension, meanwhile the new proposed algorithm is expanded from 2-dimension to 3-dimension.At the end of this paper, the proposed algorithm for non-rigid medical image registration is implemented in C++ using ITK software packages and real human liver images for the input. Comparing to viscous fluid registration method based on the gray, experimental results show that the proposed algorithm can efficiently improve the image registration accuracy, especially for large deformation of complex structures.