Aiming at the fusion problem of long-exposure image and short-exposure image under low-light condition, a patchmatch-based method is proposed to fuse low-light image pair into a better one. Firstly, a bi-directional ghost detection scheme based on pyramid multi-scale algorithm is proposed, which reduces the influence of noise and under-exposure on ghost detection, and is used to accurately detect the inconsistent region caused by the motion of the object between the image pairs.Then, the patchmatch algorithm is used to reconstruct the ghost region. And finally the Poisson fusion algorithm of improved edge characteristics is used to extract the information from the two exposures. Experimental results show that the proposed method effectively preserves sharp outlines of the short-exposed image, and maintains the color, brightness and details of the long-exposed image.