How to recognize the target from blurred image is one of the fundamental problems in computer vision and pattern recognition. Image blurring caused by abnormal energy accumulation during exposure time because of relative rotational motion between imaging system and objects. Our work is different from most of others which need “deblurring”. We try to find the invariant features between original image and blurred image based on the mathematical model of blurring and the theory of moment invariants, instead of restoring the blurred image. Method In this study, based on degraded model of rotation motion blur and Gaussian Hermit moment, we demonstrated how the rotation motion blur Gaussian Hermit moment been built and proved the existing of low-rank rotation motion blur Gaussian Hermit moment invariants. Correspondingly, rotational motion blur invariants based on Gaussian Hermit moment is built. We filtrated 5 Gaussian Hermit moment invariants from exiting rotation geometry moment invariants which had been extended to Gaussian Hermit moment invariants to construct a highly stable 5-dimensional feature vector and named it RMB_GHMI-5, and we verified that RMB_GHMI-5 had great properties of invariability and distinguishability through experiments. Finally, we introduced RMB_GHMI-5 to the field of image retrieval. Result In invariance experiment, We validate the properties of invariance of the proposed feature vector on the dataset USC-SIPI. Tow set of 18 composite blurred image and been made to test RMB_GHMI-5. The result shows the feature distance between original image and composited blurred image are extremely tiny which means RMB_GHMI-5 has great properties of invariance. In addition to the image retrieval experiments, we introduce two image database including Flavia and Butterfly for original image. Composited image which are blurred by different degree of rotation、rotational motion and Gaussian noise or Salt-pepper noise have been used to validate the invariability and distinguishability of RMB_GHMI-5. Compared with 4 state-of-the-art saliency approaches, for leaf image degraded by rotation, rotational motion and Gaussian noise, at 80% recall rate, the recognition accuracy of RMB_GHMI-5 is 25.89% higher than others. For leaf image degraded by rotation, rotational motion and Salt-pepper noise, the recognition accuracy of RMB_GHMI-5 is 39.95% higher than others. For butterfly image degraded by rotation, rotational motion and Gaussian noise, at 80% recall rate, the recognition accuracy of RMB_GHMI-5 is 7.18% higher than others. For leaf image degraded by rotation, rotational motion and Salt-pepper noise, the recognition accuracy of RMB_GHMI-5 is 3.04% higher than others. Conclusion In this study, we proposed a highly stable 5-dimensional feature vector RMB_GHMI-5, and we verified that RMB_GHMI-5 had great properties of invariability and distinguishability through experiments. The experiment results show that RMB_GHMI-5 outperforms several state-of-the-art saliency approaches and has stronger practical application value.