Study On Medical Image Registration Based On Wavelet Analysis

Along with the development of information technology, all kinds of medical imaging e-quipments were applied into the clinical medicine field. Medical images have been become the important auxiliary tools for diagnosis and treatments of diseases. And the technology of clinical diagnosis and treatments has been improved effectively. Studies on the medical image processing have been become the frontier research field and one of hot studies. In all of studies on the medical image processing, the registration of medical images was one of very impor-tant research directions. And the enhancement of the medical image was also the foundation of medical images processing and their applications. In this dissertation, we concerned with the registration and enhancement processing of medical images, and have done some researches with multi-scale analysis on the precision, the quantity of data of registration and the quality of enhancement, and so on. The main work could be summarized as follows:(1) We presented two methods for the enhancement of medical images based on multi-wavelet transform. In the first algorithm, the medical image to be enhanced was decomposed with wavelet transform, and all high-frequency coefficients were decomposed by Haar transform again. Then all high-frequency coefficients were enhanced by different enhancement weight value after they were de-noised by soft-threshold method. In the following, the image’s gray was transformed by the piecewise linear transformation after it was reconstructed. Thus the enhanced image was obtained. In the second algorithm, both high-frequency coefficients of multiwavelet transform and high-frequency coefficients of single wavelet transform were all enhanced by different enhancement weight values. In addition, all of low-frequency coefficients in every decomposition level were enhanced by a non-linear operator. Experiments have showed that the proposed methods can not only enhance an image’s details but also hold its edge features effectively.(2) We presented two methods for the rigid registration of medical images based on multi-scale analysis, which all consisted of two main procedures, i.e. the rough registration and the fine registration. In the fine registration procedure of the first algorithm, the couple medical images were decomposed by the simplified multiwavelet transform. And only the high fre-quency coefficients in the horizontal directions were selected as registration objects. Then the registration process was started from the coarse scale, and ended to the fine scale based on the images’high frequency coefficients and the rough registration information. At last, the couple initial images were selected as registration objects to accomplish the last registration based on all the above registration information. In the fine registration procedure of the second algorith-m, down-sample images were obtained by a sampling operator level by level. Then they were registered from the last level down-sample image to the initial image. In the rough registration procedure of the couple algorithms, contour features of images to be registered were extracted at first, then the principal axes method was used to obtain the translation and rotation infor-mation based on images’contour features. Experiments have showed that they were effective and accurate registration methods of medical images. Furthermore, the results demonstrated the accurate registration under noisy environment too.(3) We presented a successive approximation registration method based on the thin-plate s-plines for non-rigid medical images registration. The regions of interesting in the couple images were extracted at first, and the control points of the interesting region were selected automati-cally. Secondly, the successive approximation registration method was used to accomplish the non-rigid medical images registration, i.e., the local regions of the couple images were registered roughly based on the thin-plate splines, then, the current rough registration results were selected as the objects to be registered in the following registration procedure. Experiments showed that the proposed method was effective in the registration process of the non-rigid medical images.