An improved approach for generating globally consistent seamline networks for aerial image mosaicking
Jianan He, Mingwei Sun, Qi Chen, Zuxun Zhang · International Journal of Remote Sensing · 2018
Seamless image mosaicking is a crucial technique in the production of orthophotos in photogrammetry and remote sensing. There are many studies focused on generating high-quality ortho-mosaics and many algorithms have been put forward for this purpose. Some of the existing methods concentrate on guiding the seamline to bypass ground objects, while others focus on generating a seamline network with better global quality, based on a regular and consistent initial network. However, the former may result in an irregular seamline network structure and the latter cannot guarantee the seamline bypasses ground objects in some cases, due to limited optimization. In this study, an improved image mosaicking approach, taking both the local optimization of a single seamline and the global regularity and consistency of the whole seamline network into consideration, is proposed in order to generate a globally consistent seamline network for aerial images. Firstly, the seamline is selected using a coarse-to-fine strategy between adjacent overlapping orthoimages, in which the initial seamline is generated based on our previous work and further refined based on texture difference. Then, an effective mosaic polygon of each orthoimage is obtained through a novel method, before being used to compose a globally consistent seamline network. Furthermore, an evaluation index called the Area Weighted Distance is introduced to quantitatively evaluate the global consistency and regularity of the generated seamline network. Experiment results demonstrate that the proposed approach is feasible and promising in terms of its performance. Compared to the comparison methods, the proposed approach is 1.35 and 2.15 times smaller in the average and standard deviation of the area weighted distance, respectively, and 3.6–6.5 times faster in terms of efficiency.