In order to reduce the influence of noise on image registration accuracy, an improved image anti noise registration algorithm based on high-order cumulant is proposed. Firstly, the image is deal with average value subtraction, and the third-order spectrum is obtained by using the discrete Fourier transform of the third-order cumulant, and then the normalized cross bispectrum is calculated to effectively suppress the influence of noise. Finally, the precise translation, rotation angle and scaling factor are obtained by inverse Fourier transform. The simulation results show that the improved algorithm can achieve more accurate registration of images with strong noise, and the maximum estimation errors of translation, rotation angle and scaling are only 0.58 pixel, 0.19° and 0.36% respectively, and the registered image has smaller root mean square error and large normalization peak correlation energy confidence, which further shows that the proposed algorithm has stronger adaptability and robustness to noise, so as to obtain more accurate registration image.