To resolve the image shift problem of visible broadband spectral imaging in the field of color and imaging science, an improved image registration method based on Scale-invariant feature transforms (SIFT) algorithm is proposed. Firstly, the feature points are extracted by the SIFT algorithm. Secondly the feature points are matched by the nearest-neighbor rule of the k-d tree, and the euclidean distance is adopted to delete the false matching points. Finally, a uniform sampling method is applied to resolve the problem of SIFT feature points easy to gather and the optimal registration result is obtained. In the visible broadband spectral imaging registration test, the optimal registration result based on the uniform sampling method is compared to the result without sampling processing match points set, the former mutual information value is improved remarkably.