Star check is an important method of alignment in many multi-tracker systems. During star check, the angle error of the trackers can be calculated by comparing tracker's position when the optical axes are made parallel by sighting on a celestial body. Traditional star check has several disadvantages, such as, operating manually, spending more time, and having lower accuracy. In this paper, using mutual information as similarity measure and exhaustive method to search registration parameter, an algorithm of image registration is proposed to abstract the angle error. The results of the experiments indicate that the algorithm is very satisfactory, which can improve the efficiency of star check and almost obtain the angle error without error.