Image registration is the premise for extracting accurate inner element distribution information of samples from dual-energy computed tomographic (CT) images. In order to solve the misalignment of dualenergy CT images in synchrotron radiation, an image registration method is proposed which uses mutual information as the standard to measure similarity of two dual-energy CT images. The partial volume (PV) interpolation algorithm is used to calculate mutual information and the genetic algorithm is used to search the initial good parameters. Then the original searching point for Powell algorithm is initialized with these parameters and the PV interpolation based on Hanning window (HPV) is combined to search the optimal registration parameters. The numerical simulation testifies the advantages of HPV algorithm and the accuracy of registration achieves subpixel. Experimental results show that the proposed method is effective for dualenergy CT image registration and acquirement of three-dimensional distribution of elements.