PatchMove: Patch-based Fast Image Interpolation with Greedy Bidirectional Correspondence

Abstract

In this paper, we present a method for the plausible interpolation of images. This method has several applications, such as for smooth view interpolation, low frame-rate video upsampling, and animation. The central idea is to quickly form dense correspondences using a patch-based nearest-neighbor search method called PatchMatch. However, the conventional PatchMatch method does not always find an accurate correspondence. This means that some patches do not find appropriate counterparts. Our method employs a greedy algorithm and an occlusion handling technique to correct inaccurate correspondences. Furthermore, our texture reconstruction method successfully reduces blurring effects. We demonstrate that our method significantly reduces the computation time required for interpolation, and show that the quality of reconstructed images is almost identical to that of those generated using state-of-the-art methods.