SEAMLINE OPTIMISATION FOR URBAN AERIAL ORTHO-IMAGE MOSAICKING USING GRAPH CUTS

摘要

Optimal seamline detection is a key step when composing digital orthophotomaps (DOM) of extensive areas from overlapping ortho-images. To avoid seamlines passing through buildings when mosaicking, seamline detection between adjacent ortho-images is casted as a graph-cut problem. To avoid buildings, colour differences and gradient magnitude derived from the ortho-images, together with approximate object heights derived from a digital surface model (DSM), are combined for constructing the cost term of the graph. To efficiently obtain an approximate height value for each DOM unit, an expanded normalised DSM is proposed. Furthermore, the designed cost term for final composite ortho-image merging is weighted to prioritise pixels closest to the nadir point. Experiments with strips of aerial images show that the proposed method obtains seamless mosaicking of ortho-images without manual work.