An image registration method based on complex Zernike moment phase angle estimation was proposed. Firstly, the Harris-laplace operator was used to detect interest points in an image, and the interest points were regarded as initial feature points. The Zernike moments defined on the scale normalized interest point neighborhood were computed, and a new robust estimation method for phases was presented to compute the rotation angle between two normalized regions. Then, the magnitude and phase angle information of Zernike moments were combined and used to measure the Euclidean distance between two matching regions. Finally, an iterative refined angle method was proposed to estimate the parameters accurately, and the image registration was finished after the geometric transform of input images. The experimental results show that the proposed algorithm impletments a precise image registration under the scaling, arbitrary rotation and noise. The average coverage percentage achieves 94.125% when the rotation angle error is less than 20°, which reduces the false match rate effectively.