We present a novel approach to boost image matching performance by fusing multiple local descriptors in the homography space. Traditional matching methods find correspondences based on a single descriptor and the performance becomes unstable due to the goodness of the chosen descriptor To address this problem, our method uses multiple descriptors and select a good descriptor for matching each feature point. Specifically, we project every correspondence into the homography space, where correct correspondences tend to gather together due to the similarity of their homographies. Then kernel density estimation is applied to measure the density in the homography space and verify the correctness of correspondences. The proposed approach is comprehensively compared with the state-of-the-art methods and the promising results manifest its effectiveness.