Geometric correction of very high resolution satellite images using virtual sensor model

In this paper, a geometric correction architecture for very high resolution (VHR) satellite images acquired by push-broom imagers is proposed. The idea is based on an ideal virtual imager mounted on a virtual perfect platform that would produce flawless images with no geometric errors whatsoever. The idea is to generate virtual attitude and orbit data from the actual measurements by fitting appropriate models. Based on this data, it is possible to geo-reference this virtual image through geo-location process. Once a geo-referenced perfect image geometry is obtained, the rest is to fill in its pixel intensity values by resampling from the geometrically raw image. In order to relate the raw geometry to virtual perfect geometry, the raw image has to be geo-referenced as well. By doing so, the two geometries can be linked by earth coordinates. This method is inspired by a method known as perfect sensor which is commercially used but poorly described in the literature.