Orthorectification is the process of removing sensor (scanner or camera), satellite/aircraft, and terrain-related distortions for creating a planimetrically correct image.
To obtain an accurately orthorectified image, the following information is required: (1) accurate elevation model, and (2) a camera model or rational polynomial coefficients (RPCs) that depicts the positional relationship of the collected image to the ground. Many companies deliver their images together with RPCs and existing software implementations can automatically read these files and apply the RPC transformation on the fly. An accurate elevation model is important to remove the influence of topography (e.g. hills, valley, etc.) on the raw image so that users can accurately compute distances, areas, and directions. Without performing orthorectification, the features in the image are tilted (especially the features located away from the center of the camera). Many satellite data products (e.g. Sentinel images, Landsat data products) are orthorectified using Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) data which is a freely available data product and has a spatial resolution of e.g. 1 arc-second (30 m). In the case of extremely jagged surface topography, i.e. areas of high relief, a DEM with a higher spatial resolution is required.
Two main models can be used in the orthorectification process: black-box and the physical-based model. The black-box model (called also the analytical model) is commonly implemented in different software because it relies solely on the RPC files. This model does not require access to any proprietary information of the sensor used to collect the image.
The physical-based models are more complex (and hence expected to be more accurate) because they account for various factors that might influence the quality of the acquired image: e.g. position of the satellite when collecting the images, atmospheric effects, etc. An example of a physical-based model is the so-called camera model. This model requires access to proprietary sensor information that has to be provided by the image owner.