Research On Registration And Fusion Of Medical Image

In recent years, the rapid development of modern medicine promotes the extensive application of medical image. The different information of relevant organs are reflected on the different medical images, complement each other, which become the important reference of clinical diagnosis and treatment. In clinic, in order to obtain more accurate image information organization of lesions, people often need to comprehensively analyze a variety of medical images. As a result, the medical image fusion is particularly important. And it is supposed to obtain more reliable and comprehensive results by reducing the redundant information of image fusion and adding complementary information. And, the medical image registration is the precondition of fusion, which would directly affect the final fused result.The article introduces the significance of medical image registration and the recent development situation, and analyzes the basic principle of registration and specific implementation process, and study the medical image registration algorithm based on mutual information. Besides, the influences of the interpolation method and the search algorithm of mutual information registration algorithm are further discussed. Due to the problem of mutual information easily to fall into local extreme in medical image registration method, the article proposes that the partial volume (PV) interpolation is modified and the gradient information of image is integrated in the mutual information to improve the accuracy of image. In order to improve the registration speed, this paper adopts method of moment as the method of the initial registration, which reduces the consumption of time. The experimental results show that the improved registration method can effectively realize the medical image registration.After registration, medical image fusion is finished based on the PLIP (Parameterized of Logarithmic Image Processing) model and discrete wavelet transform method. With defining a set of parameterized nonlinear operation to alternative linear operation of image, the grey value is directly used in the PLIP model, which describes more accurately the nonlinear of image processing. The PLIP model is suitable for signal processing, especially for image enhancement, edge detection and image restoration. Discrete wavelet transform has the character of focusing on the time-frequency. The different resolution and multiple directional wavelet sub-bands can be got by the wavelet transform of the processed image under the PLIP algorithm, which provide the possibility of image fusion with different sub-bands. According to wavelet sub-band coefficients fusion rules, a new method is proposed in this paper, which based on the weight of the human visual selection method and regional variance consistency check with adding window selection method, respectively, as to the low frequency sub-band and the high frequency sub-band coefficients. Then the fused image is obtained by the inverse wavelet transformation and the PLIP model inverse operation. Experiments prove that the proposed image fusion method based on PLIP model and discrete wavelet transform is good, which could meet the demands of the visibility of human eyes.