Mosaicing and Frames Registration








The creation of a mosaic from a sequence or a collection of images has been popular in recent years due to its large number of applications. Most of the work have focused on the estimation of the parameters modeling the camera motion from a pair of images. The obtained results are insufficient to generate a seamless mosaic of a complete scene. Although the pairwise registration is accurate, the concatenation of these pairwise transformations leads to global alignment errors. Therefore, the use of a global registration technique is required to build a consistent large view from a collection of images.

We propose a graph-based approach merging local and global registrations into a single framework. The graph representation of the topology of the swaths allows us to search for the optimal path connecting every frame in the sequence to the considered reference frame. This search is based on a cost function associated to each arc of the graph. This cost reflects the pairwise registration accuracy based on image correlation and a grid deformation representing the set of feature points selected in the mosaic projection space.




Original soccer sequence Computed mosaic
Original basketball sequence Computed mosaic