SIFT registration can obtain scale invariant and rotation invariant features easily. SIFT requires images with rich details, and the amount of calculation is too large. SIFT for infrared image and visible image registration, discarded the edge features and the low contrast pixels directly. Especially when the image with low contrast in sea and sky background, SIFT registration cannot get enough feature points and desirable effects. The classic SIFT algorithm often result in feature point degradation. A novel approach has been proposed, which through the most stable extremely region flock the SIFT feature points. Analysis the worth of low contrast and edge information in sea and sky background image, use it to rough registration. Comparison of the algorithm with an algorithm of the same kind, the results show the effectiveness of the registration with low ergodic comparisons.