Study Of Medicinal Image-rigid Registration Methods Based On Flexible Model Of United Two Features

By implementing spatial geometric transformation on two or more images of different times, different fields, or different imaging patterns, the main purpose of Medicinal Image Registration is to make sure that the pixels represent the same anatomical structure or body can match elements in the geometry correspond. In this process, it is of great importance to pay attention to all the indicators as high degree of automation, robustness, adaptability, speed, precision indicators.Feature-based image registration method is one of the most commonly used image registration methods. Its biggest advantage is its ability in transforming the analysis of a whole image into the analysis of a variety of image features (feature points, curves, etc.). In this way, it greatly reduced computation in image processing. What’s more, the feature extraction accuracy and positioning accuracy throughout the registration process would have a huge impact on the whole process. This paper analyzed the existing features of extraction methods with special attention on extraction methods of point features and contour features. It introduced a brand new way for Medicinal Image Registration, which is a kind of combination of Harris corner detection method and LOG edge detection method approach. First, it extracts image corners by employing Harris corner detection method and increases their gray value. Then, it extracts contour by employing the LOG edge detection method. Finally, all the features would have a free competition.At the same time, it studies cubic B-spline model, afterwards using cubic B-spline interpolation spline Mixed to do interpolation, and then employing Hausdorff as the measuring method. PSO method is also used to better the process of image registration. The image registration process of the measuring method used in this thesis is also discussed in this thesis.This thesis developed the registration process of OpenCV development platform, which enabled the realization of the image registration method of this essay. Many experimental data and comparative trials on images of CT. MR and PET are made under the direction of this method.It evaluated indicators such as correlation coefficient, minimum mean square error, normal mutual information and signal-noise ratio, which are all means to assess objectively the registration results.Judging from the evaluate results, method presented in this thesis is accurate and robust. can carry out many types of image registration effectively. It also shows the effectiveness and value of the outcome of this study.