Image registration for super resolution using scale invariant feature transform, belief propagation and random sampling consensus

Abstract

Accurate image registration is crucial for the effectiveness of super resolution. In super resolution, image registration is used to find the disparity between low resolution images. In this paper an image registration approach based on a combination of Scale Invariant Feature Transform (SIFT), Belief Propagation (BP) and Random Sampling Consensus (RANSAC) is proposed for super resolution. The SIFT algorithm is used to detect and extract the local features in images, BP is used to match the features while RANSAC is adopted to filter out the mismatched points and then estimate the transformation matrix. The proposed method is compared with traditional SIFT to verify its accuracy and stability. Finally, the result of using the proposed approach in the super resolution application is given.