This paper presents an automatic superresolution method using multiviewpoint images. Viewpoints of objects in consecutive image frames often change in real video sequences. Hence, in order to adapt the conventional superresolution methods to multiviewpoint images, a geometric transformation of multiviewpoint images to a reference image plane should be performed. In this process, an accurate isoplane transformation is required. We propose a robust random sample consensus (RANSAC) criterion and a weighted homography estimation, which are important for accurate geometric transformation. Experiments were performed with several video sequences that simulate real surveillance systems. Multiviewpoint image sets were also used to verify the accuracy and stability of the proposed methods. The experimental results show that low-resolution images that are difficult to discern become recognizable high-resolution images using the proposed superresolution method. (c) 2008 Society of Photo-Optical Instrumentation Engineers.