In this paper, we propose an improved version of the neighbor-based super-resolution algorithm proposed by Chang et al.. Different from Chang’s method that uses the Euclidean distance to find the K most nearest neighbors of a low-resolution patch, we define a similarity function and use it to find the K most similar neighbors of a low-resolution patch. In addition, we use a set of different weights for taking a linear combination of the high-resolution patches corresponding to the selected K most nearest neighbors. Although the set of the weights used by Chang minimizes the error they defined, the reconstructed high-resolution images by our method have better PSNR and SSIM than those constructed by Chang’s method.