The traditional method of computing rotation invariant moments of image,which needs to convert the image into the polar coordinate system,not only increases the computational load greatly,but also creates large rounding error,and results in error rate increase during image recognition and reconstructed image quality deterioration with image moments. To avoid the disadvantages caused during the process of pixel conversion,this paper took the computation of radical harmonic Fourier moments as example,and proposed a new method of computing moments in rectangular coordinate system immediately. The results of the experiments show that the new method can not only eliminate error caused by coordinate conversion,but also significantly reduce computational load.