The Research On Mutual Information Based And Discrete Wavelet Frame Decomposition Based Medical Image Registration

Thanks to the gradually maturing of information technology, modern medical imaging technology is developing at a very fast speed and getting a more and more important role in medical research and diagnosis. It is often not enough to use a single image in nowadays complicate medical applications due to the limitation of imaging equipments. Image registration arises as the first problem to be settled out when it is necessary to integrate information coming from different images. Medical image registration is getting more and more important in the field of medical image analysis, and is one of the hottest sub disciplines.Registration methods based on mutual information are being used widely, but local optimums are introduced by the Partial Volume interpolation in conventional version. Registration may thus converge to a wrong point by adopting Powell's method as the optimize tool. An expanded PV interpolation was introduced in this paper, which is based on B-spline function and is going to reduce the local optimums. An hybrid search strategy which combines genetic algorithm and Powell's method is adopted to avoid the impact of local optimum. Experiments showed that the proposed method is more stable than the conventional method and is capable of getting higher accuracy.Wavelet based methods have been widely used in remote sensing image registration, but comparatively less in medical image registration. The second contribution of this paper is a novel discrete wavelet frame transformation based registration method. Energy maps of sub-images are computed by destructing the original images, and are used as the feature to be registered by the metric of sum of absoluted distance. Genetic algorithm was adopted to get the minimum of the registration function. Experiments illustrated the effectiveness of the newly proposed algorithm, and better accuracy in X axis was achieved in rigid transformation compared to mutual information method, normalized mutual information method and correlation ratio method.