Abstract: In order to solve the problems like low corresponding points matching rate，poor automatic registration process of multi-source optical satellite imagery in geometric correction. We described the principal direction and feature vector by phase congruency instead of the original SIFT algorithm gray level of pixel gradient to ascending accuracy of feature description.Optimizing the matching results with constraints of phase analysis results under the frequency domain，inhibition of error matching points;meanwhile，put forward a adaptive selection strategy to improve parameter estimation phase level of automation RANSAC-based;the ultimate realization of automatic image registration between the multi-source optical satellite imagery. The experimental results of multiple sets of data show that the validity and applicability of this schema to image registration between multi-source optical satellite image with radiation and nonlinear distortion.