Issue No. 04 - April (2008 vol. 30)
Jiaya Jia , IEEE
Chi-Keung Tang , IEEE
The aim of this paper is to achieve seamless imagestitching without producing visual artifact caused by severeintensity discrepancy and structure misalignment, given that theinput images are roughly aligned or globally registered. Our newapproach is based on structure deformation and propagation forachieving overall consistency in image structure and intensity.The new stitching algorithm, which has found applications inimage compositing, image blending, and intensity correction,consists of the following main processes. Depending on thecompatibility and distinctiveness of the 2-D features detected inthe image plane, single or double optimal partitions are computedsubject to the constraints of intensity coherence and structurecontinuity. Afterwards, specific 1-D features are detected alongthe computed optimal partitions, from which a set of sparsedeformation vectors is derived to encode 1-D feature matchingbetween the partitions. These sparse deformation cues are robustlypropagated into the input images by solving the associatedminimization problem in gradient domain, thus providing auniform framework for the simultaneous alignment of imagestructure and intensity. We present results in general imagecompositing and blending, in order to show the effectiveness ofour method in producing seamless stitching results from complexinput images.
Image stitching, structure deformation, image alignment
C. Tang and J. Jia, "Image Stitching Using Structure Deformation," in IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 30, no. , pp. 617-631, 2007.