Image mosaic is the key step to panorama

while existing algorithms focus on scenes with the same illumination. This paper presents a novel method for image mosaics with notable illumination difference. First

feature points are extracted by ring projection

which overpass the local area limitation in tradition methods and gains exactly matching result for image under vary illumination conditions. Then mean and covariance are used to adjust the global brightness of the images to solve the problem of different illumination. While blending factors are gained using the energy function that contains image gradient

which conquers the blur and mackle problems in the overlap area if using the tradition linear weight function. Experiment results show that we can gain satisfying visual effect using this method in image mosaics with notable illumination difference.