Optical Co-registration

Every optical image arrives georeferenced and almost none arrive perfectly co-registered. Residual orbit and attitude errors, terrain parallax on off-nadir acquisitions, different elevation models used in orthorectification, and successive processing baselines all leave a shift behind, typically a fraction of a pixel, often several. For a mosaic that is cosmetic. For change detection it is fatal: a one-pixel shift along a field boundary or a roofline produces a bright edge of false change that no threshold can separate from the real thing.

This service removes that shift, whatever produced the image. It aligns each scene to a chosen reference, an external reference dataset where one exists for that sensor, a master scene from the stack, a stack median, or the customer’s own orthophoto or cadastral base, by matching stable features and estimating the offset to sub-pixel precision. Where a single global shift is not enough, because terrain parallax or a strongly off-nadir geometry distorts the scene locally, it fits a local warp field instead. Matching works across sensors and across resolutions, so a metre-class commercial acquisition, a Sentinel-2 scene, a drone orthophoto and a decades-old scanned archive frame can be brought onto the same grid as the open medium-resolution archive.

The output carries a quality report stating the residual shift actually achieved, per scene, in metres and in pixels, so a downstream analyst knows whether a detected change is larger than the registration error. That number is the difference between a result and a claim. It is also what makes a genuinely multi-sensor time series possible: one coherent record spanning whatever imagery the customer can get, rather than several incompatible ones.