Patch-based Fast Image Interpolation in Spatial and Temporal Direction
Shunsuke Saito (Waseda University), Ryuuki Sakamoto, Shigeo Morishima (Waseda University)
The 41st international Conference and Exhibition on Computer Graphics and Interactive Techniques (SIGGRAPH2014) Posters, 2014/8
画像処理 (Image Processing)
- Recently, a plausible interpolation of images has been proposed for many applications such as smooth view interpolation, up-sampling low frame rate video, and making animation from still images. However, producing an intermediate frame from two images is still a challenging task. The moving gradient method enables one to obtain intermediate frames without suffering from blurring or ghost effects, by computing the displacement of each pixel [Mahajan et al 2009]. However, their method requires many iterations of the optimization process for pixel displacements by graph-cut. Hence their method is not suitable for interactive applications. To compute correspondence between two images rapidly, we employ the PatchMatch framework [Bames et al 2009], which is a significantly faster method for reconstructing an image from plenty of patches. However, an issue is that it is not necessary to find accurate correspondence in this method, which means that while some patches find appropriate counterparts, the others do not. We employ a greedy approach so that the incorrect correspondences can follow the accurate ones. Finally, we show that significant improvements on computing times are obtained, and that difference of the results between the state-of-the-art approach and our method is negligibly small.