This paper proposes a fast technique for matching a query image to numerous database images under geometric variations in rotation, scale, and translation. Our proposed method extracts the Fourier-Mellin phase features from the images for invariant matching. The online matching process in our method is fast because it directly determines identification based on the correlation value between those features without the geometric alignment. Our method also proposes selecting only the low frequency bands of those features to be matched for improving identification accuracies, as the high frequency bands tend to be inaccurate in Fourier-Mellin transform. The benefits of our method were verified in our experiments to match the “fingerprint” images of 1,210 metal shafts. The processing time of the 1-vs.-1,210 matching was only 0.136 seconds on a standard desktop computer, and all individuals were identified with no errors.