FSIS: Fast and Seamless Image Stitching of High-Resolution Dunhuang Murals

Abstract

Digital processing of Dunhuang murals is important for the permanent and real preservation of precious cultural relics information. The image stitching is one of the key and challenge steps to obtain complete murals with high resolution. Gradient-domain methods, which need to solve a sparse linear system of the Poisson equation, are widely used in the field of image stitching. However, it is a time consuming and resource-intensive task for solving the Poisson equations, especially when the image is large. To address the above problems, a fast and seamless image stitching method called FSIS implemented on graphics processing unit (GPU) architecture is proposed to improve the image stitching efficiency of high-resolution Dunhuang murals. Specially, the original image is firstly down-sampled into low-resolution border pixels images (BPIs) by using quadrilateral grids. Next, the Poisson equations are constructed and solved based on the BPIs. Finally, the final result image is obtained by interpolation. Through using the above steps, the time and memory consumed by solving Poisson equations can be reduced. Experimental results have shown the efficiency of the proposed method.