Abstract: An image mosaic method based on SIFT and Selection Region is described in this paper.It uses scale invariant feature transform (SIFT) to extract invariant features from images mosaic, makes full use of the rotational invariance of circles and the maximum mutual information for feature points matching, this method can avoid a heavy computing load and the inaccurate matching of feature points compared with the traditional registration algorithms, and weighted average method is used to blend images.Experimental results demonstrate that the proposed method is robust to translation, rotate, intensity and noises, and can produce high quality image mosaic.