Large Parallax Image Stitching via Structure Preservation and Multi-matching

Abstract

AbstractIn image stitching tasks, parallax ghosts and composite ghosts widely exist in stitching results whose input images usually have complex moving objects (especially multiple pedestrians) and large parallax scenes, which bring greater challenges to image stitching. In order to efficiently remove these ghosts and maintain visually natural stitching results, we propose a novel image stitching method via structure preservation and multi-matching (SPMM). Specifically, the method is composed of the structure preservation module based on grid constraints that eliminates parallax ghosts by ensuring the alignment of overlapping regions and keeping object structures in non-overlapping regions; and the composite ghost removal module based on Multi-Matching, which effectively eliminates composite ghosts caused by multiple pedestrians and obtains natural stitching results. The extensive experiments on five different image stitching frameworks and real datasets demonstrate the effectiveness and advantages of the proposed method. KeywordsImage stitchingStructure preservationMulti-matching