The work aims to solve the defects such as low registration accuracy and poor efficiency of current image registration algorithm due to the application of l1 distance or l2 distance similarity measuring means to complete image feature matching, causing it to ignore the phase information and making it hard to effectively eliminate the effect of Gaussian noise. The anti-noise image registration algorithm based on optimal similarity distance coupled with radial angle transformation was proposed. Firstly, image feature point was fast extracted by introducing the radial angle transformation to reduce the complexity of the algorithm; then the optimal similarity distance measurement model was defined by combining the image amplitude and phase information and based on the measurement of Euclidean distance to improve the anti-noise performance to complete the feature points matching through solving its global minimum. Finally, the image was divided into inner points and outer points. 6 inner points were selected to improve the consistent strategies of random samples to purify the matching and eliminate the mismatching through the calculation of their geometrical registration errors of transformation matrix. Simulation results showed that: this algorithm had stronger anti Gaussian noise performance and higher matching accuracy, as well as the shortest consumption of time compared with the current image registration techniques measured based on l1 distance or l2 distance similarity. The algorithm mentioned can accurately com-plete the image feature registration.