Hybrid Network-Based Automatic Seamline Detection for Orthophoto Mosaicking.

Seamline detection is a crucial procedure for orthophoto mosaicking. To eliminate the seam effect caused by geometric discontinuities, seamlines must avoid crossing areas containing the obvious ground object, for which manual processing is usually required. Many existing seamline detection methods can generate seamlines bypassing most of obvious ground objects but always take pixel-level computation which may consume much time. To address this problem, this paper presents a seamline detection approach based on a hybrid network search. First, without auxiliary data, the semi-global block matching (SGBM) algorithm was adopted to generate a disparity map for pairwise orthophoto overlapping area. By using adaptive threshold segmentation, a binary cost map containing the ground objects was obtained. Subsequently, a hybrid network was constructed by edge points and uniform points extracted on the cost map. Finally, seamlines were detected by search on this network-based graph. The essential contribution of the proposed method is that the seamline is searched on a sparse hybrid network instead of a raster cost map. Thus, computational complexity can be significantly decreased and produce fine-tuned seamlines. A Series of comparison experiments were carried out between the proposed and well-established methods, using two benchmark datasets with different characteristics. The comparison results demonstrated that the proposed method could generate high-quality seamlines in terms of visual comparison and statistical evaluation. Moreover, the processing speed has a nearly tenfold improvement compared with the control group methods.