Image Stitching and Blending of Dunhuang Murals Based on Image Pyramid

Abstract

It is a feasible and meaningful way to preserve the precious murals by using digitization technology. However, it may need to stitch multiple images to get a complete mural panorama. It poses great challenges to stitch and blend these multiple images in a seamless and visually pleasant way. In this paper, we propose an image stitching and blending method based on image pyramid, including Gaussian and Laplacian pyramid, for obtaining a natural stitched images of Dunhuang murals. In special, Gaussian pyramid is used to down-sample the images and get multiple images of different scales through using Gaussian filters. Laplacian pyramid is applied to generate the difference image between each levels of the Gaussian pyramid. Finally, the images can be blended in an adjustable way by specifying the weights of different images. In this way, the stitched images can be presented with reasonable image fusion. Through using the real Dunhuang murals images, the efficiency of the proposed method has been validated.