Abstract: Aiming at the resampling and requantization error caused by image rotation and scaling transformation, the improved Zernike moment method was proposed, that is, first the target area in the image was normalized shapely, and then the Zernike moments were normalized. The experimental data showed that the improved Zernike moments not only had rotation invariance, but also had the scale invariance which didn't have before improved. The classification results of the target to be identified based on the minimum distance classifier showed that the improved Zernike moments had a higher recognition rate. Its shortcomings are not applicable to the complex situations of target image background.