In multi-modal images (MMI), the differences in their imaging mechanisms lead to large signal-to-noise ratio differences, which means that the matching of geometric invariance and the matching accuracy of the matching algorithms often cannot be balanced. Therefore, how to weaken the signal-to-noise interference of MMI, maintain good scale and rotation invariance, and obtain high-precision matching correspondences becomes a challenge for multimodal remote sensing image matching. Based on this, a lightweight MMI alignment of the phase exponent of the differences in the Gaussian pyramid (PEDoG) is proposed, which takes into account the phase exponent differences of the Gaussian pyramid with normalized filtration, i.e., it achieves the high-precision identification of matching correspondences points while maintaining the geometric invariance of multi-modal matching. The proposed PEDoG method consists of three main parts, introducing the phase consistency model into the differential Gaussian pyramid to construct a new phase index. Then, three types of MMI (multi-temporal image, infrared–optical image, and map–optical image) are selected as the experimental datasets and compared with the advanced matching methods, and the results show that the NCM (number of correct matches) of the PEDoG method displays a minimum improvement of 3.3 times compared with the other methods, and the average RMSE (root mean square error) is 1.69 pixels, which is the lowest value among all the matching methods. Finally, the alignment results of the image are shown in the tessellated mosaic mode, which shows that the feature edges of the image are connected consistently without interlacing and artifacts. It can be seen that the proposed PEDoG method can realize high-precision alignment while taking geometric invariance into account.