Research Of Medical Image Registration Algorithm Based On Canny Operator Combined Wavelet Transform And Improved Hausdorff Distance

Medical image registration is an important aspect of the application of modern medical image processing technology. Its main target is taking spatial geometric transform to two or more images obtained at different times, from different views, or of different imaging modes, to make all the anatomical points of the images, or the pixels or voxels that represent the same anatomical structure match geometrically. This thesis takes research on the subject.This thesis analyzes the existing feature extraction methods, focuses on the extraction methods of the image edge feature. Aiming at the disadvantages of Canny operator and wavelet transform in edge detection, a new method for image edge detection has been proposed, that is the Canny operator combined wavelet transform based edge detection algorithm. The method first extracts the image edge using Canny operator and wavelet transform respectively, then eliminates the noise in the edge extracted by Canny operator using the edge extracted by wavelet transform, finally further processes the edge and obtains the final edge image.Then the paper studies the Hausdorff distance metric. The algorithm computes simply, and does not need the one-to-one corresponding relation among the feature points. However, Hausdorff distance only contains part of the image information, cannot resolve the situations that the object is covered and there are big shape changes, and it is sensitive to noise. Considering these problems, the difference squared Hausdorff distance metric has been proposed, adding the difference square between the images to the original Hausdorff distance.The registration algorithms used have been integrated in the paper, with the transform model using the high quality B spline elastic transform, using the simulated annealing combined genetic algorithm to optimize the transforming parameters, avoiding the registration result trapped into local minimum. The complete process of the non-rigid registration algorithm has been given. A large number of registration experiments have been carried out on OpenCV platform among single and multi-mode CT, MR, PET images. Comparions among the results using the algorithm in the paper, the Canny operator and the traditional Hausdorff distance algorithm have been carried out. The results show that the algorithm in this thesis is superior.