Image provider agencies deliver images with standardized dimensions. For example, the dimensions of a single WorldView-2 image tile are 16.4 x 16.4 km. Thus, if the dimensions of your area of interest are bigger than these, your image data will be delivered in multiple, adjacent tiles. In such cases, the first processing step is to mosaic the tiles into one seamless image.
Image mosaicking is performed on the basis of a common reference/coordinate system that is reported in the metadata (Turner et al., 2014, 2012). This system may be a real world coordinate system (i.e., a map projection) or simply a grid that defines the position of pixels within the image. Whether one can choose between the two depends on the information that is in the metadata, which to a large extent also depends on the processing level of your original product. Sometimes, the coordinates that come with the images are not accurate, although they permit mosaicking of adjacent tiles. This is the case for the level 2A products from DigitalGlobe, which were delivered to the STARS project. Even in cases when the data does not come with a coordinate system, it is possible to mosaic adjacent tiles on the basis of pixel positions of the upper-left corner of each tile (which are provided in the metadata).
Mosaicking can be performed with almost all proprietary and open-source remote sensing applications. Popular open-source applications such as QGIS, ILWIS, MultiSpec and the Orfeo Toolbox can perform image mosaicking. In the STARS automated processing workflow, a script based on GDAL was developed for this purpose. The script extracts reference information from the metadata to this end. In cases of overlap between image tiles, the script assigns the average value of the overlapping pixels to the output image.