To solve the problem of the extensive feature-matching time of the grid motion statistics (GMS) algorithm when the image is rotated, an improved fast GMS image matching algorithm is proposed herein. First, a fast feature point extraction and description algorithm is used to violently match the image to obtain the initial matching point pair. The feature matching pair with a higher matching quality is selected according to the Hamming distance. Second, the approximate rotation angle of the two images is obtained by calculating the difference between the main directions of the matching and the feature points. The best form of the motion core is directly determined according to the rotation angle. Finally, the GMS algorithm is used to distinguish between correct and wrong matching points according to the motion core, avoiding the loop calculation of the matching results of different forms of motion cores and reducing the running time of the algorithm. The experimental results show that the number of correct matching points of the improved algorithm is at least 28% higher than that of the traditional feature point matching algorithm, and the matching speed can be at least 36% higher than that of the GMS algorithm.