Contrast-enhanced multiphase liver CT is currently a standard of practice in hepatic imaging for medical diagnosis of liver diseases. The main contribution of this paper is an image registration method which finds an accurate alignment of arterial phase and portal venous phase liver CT images. The proposed method combines global rigid registration (affine) and local nonrigid deformable registration (B-spline) in a manner of alternative iterations, and extends traditional two-step registration to iterative multi-step registration. During the registration, rigid and nonrigid registrations compensate each other by updating the transformed coordinates successively. Meanwhile, a composite optimization strategy is employed to tradeoff between the convergence and computational complexity. In our experiments, we derived the semi-automatic controlled transformations from registering large volume of data and in return used these transformations to evaluate the accuracy. Furthermore, experimental comparisons of our method with two-step methods were conducted. The corresponding results showed the proposed method is a promising scheme for image registration in terms of accuracy and robustness.