This paper introduces a fast and simple geometric solution for solving the problem of example-based image superresolution with the advantages of well suppressing noise. Here, an image is considered as a set of small image patches, and super-resolution is performed on each patche with the help of a given database of low-resolution and high-resolution image patch pairs. For each given low-resolution patch, to estimate its corresponding high-resolution patch, a set of candidate highresolution patches is first searched from the database using a criteria based on Euclidean distance and statistical properties. Then, by considering each image patch as a point in a high dimensional space, the disered high-resolution patch is determined using the projection of the low-resolution point onto the convex hull of the candidate high-resolution points. Experiments are carried out to demonstrate the performance of the proposed method over some state-of-the-art methods.