In this paper, the method of image mosaics based on feature points and color transfer is presented. First, feature points are extracted and matched by sift algorithm, probabilistic screening is used to improve the accuracy of matching. Then, in lalphabeta color model, the algorithm of color transfer is improved to adjust the brightness of the images to solve the problem of different illumination. Graph cut and multiresolution fusion techniques are used to cope with ghosting. The affectivity of the method is demonstrated on our experimental results.