This paper presents monomodality registration of rigid body, elastic rigistration of non-rigid body and interpolation used in image registration. By comparison of several interpolations, because of the best tradeoff between accuracy and the computational cost, cubic spline interpolation is selected as interpolation of this paper's registration. For global registration of rigid body, based on analysis of the principal axes method, by making use of automatic and very fast nature ,and easy implementation of the method, a registration method of principal axes-information is presented to remedy drawback of coarse result. In this method, the principal axes method is used to estimate the parameters of the transformation, then with the estimated pamameters as the initialization of Powell optimization method and with all of the available information as the feature, a high accurate registration is achieved by calculating the minimum standard deviation between the registered images. For elastic reigistration of non-rigid body, after the thin-plate splines for image registration is analysed, the local elastic registration is presented, which usilizes a positive definite function with compact support as the radial basis function of interpolation.
|