Globally consistent alignment for planar mosaicking via topology analysis

In this paper, we propose a generic framework for globally consistent alignment of images captured from approximately planar scenes via topology analysis, capable of resisting the perspective distortion meanwhile preserving the local alignment accuracy. Firstly, to estimate the topological relations of images efficiently, we search for a main chain connecting all images over a fast built similarity table of image pairs (mainly for the unordered image sequence), along which the potential overlapping pairs are incrementally detected according to the gradually recovered geometric positions and orientations. Secondly, all the sequential images are organized as a spanning tree through applying a graph algorithm on the topological graph, so as to find the optimal reference image which minimizes the total number of error propagation. Thirdly, the global alignment under topology analysis is performed in the strategy that images are initially aligned by groups via the affine model, followed by the homography refinement under the anti-perspective constraint, which manages to keep the optimal balance between aligning precision and global consistency. Finally, experimental results on two challenging aerial image sets illustrate the superiority of the proposed approach.