Panorama Weaving: Fast and Flexible Seam Processing

PanoramaWeaving on a challenging data-set (Nation, 12848 x 3821, 9 images) with moving objects during acquisition, registration issues and varying exposure. Our initial automatic solution (bottom, left) was computed in 4.6 seconds at full resolution for a result with lower seam energy than Graph Cuts. Additionally, we present a system for the interactive user exploration of the seam solution space (bottom, right), easily enabling: (a) the resolution of moving objects, (b) the hiding of registration artifacts (split pole) in low contrast areas (scooter) or (c) the fix of semantic notions for which automatic decisions can be unsatisfactory (stoplight colors are inconsistent after the automatic solve). The user editing session took only a few minutes. (top) the final, color-corrected panorama.
A fundamental step in stitching several pictures to form a larger mosaic is the computation of boundary seams that minimize the visual artifacts in the transition between images. Current seam computation algorithms use optimization methods that may be slow, sequential, memory intensive, and prone to finding suboptimal solutions related to local minima of the chosen energy function. Moreover, even when these techniques perform well, their solution may not be perceptually ideal (or even good). Such an inflexible approach does not allow the possibility of user-based improvement. This paper introduces the Panorama Weaving technique for seam creation and editing in an image mosaic. First, Panorama Weaving provides a procedure to create boundaries for panoramas that is fast, has low memory requirements and is easy to parallelize. This technique often produces seams with lower energy than the competing global technique. Second, it provides the first interactive technique for the exploration of the seam solution space. This powerful editing capability allows the user to automatically extract energy minimizing seams given a sparse set of constraints. With a variety of empirical results, we show how Panorama Weaving allows the computation and editing of a wide range of digital panoramas including unstructured configurations.
	
@article{Summa:2012:PWF:2185520.2335434,
         author = {Summa, Brian and Tierny, Julien and Pascucci, Valerio},
         title = {Panorama weaving: fast and flexible seam processing},
         journal = {ACM Trans. Graph.},
         issue_date = {July 2012},
         volume = {31},
         number = {4},
         month = jul,
         year = {2012},
         issn = {0730-0301},
         pages = {83:1--83:11},
         articleno = {83},
         numpages = {11},
         url = {http://doi.acm.org/http://doi.acm.org/10.1145/2185520.2185579},
         doi = {http://doi.acm.org/10.1145/2185520.2185579},
         acmid = {2335434},
         publisher = {ACM},
         address = {New York, NY, USA},
         }