A Parallel Multigrid Poisson PDE Solver for Gigapixel Image Editing

Abstract

With the development of image acquisition technology, gigapixel images are easily produced and widely used in modern society. How to efficiently compile these gigapixel images within gradient domain is the research focus in the community of image processing and computer graphics. To solve Poisson equations involving large-scale unknowns is crucial for gigapixel image editing in gradient domain. Traditional multigrid approach separately performs iteration, restriction and interpolation, bears heavy communication costs between RAM and external memory. In the paper, a parallel multigrid Poisson solver for gigapixel image editing is proposed, which exploits the locality and relevance of memory accessing and updating among the iteration, restriction and interpolation for parallel performing the iteration, restriction and interpolation in the sweeping window. Image stitching experiments show that the presented method exhibits the higher efficiency than the Poisson solver of successive overrelaxation, gauss-seider iteration and traditional multigrid.