Registration is the process of aligning two or more images with overlapping fields of view, captured at different times or with different sensors. Accurate alignment of images is an essential step before further image processing can take place.

AUG Signals has been a leader in the area of automatic image registration for over fifteen years, having developed leading edge technology and software capable of automatically registering multi-sensor images at a sub-pixel registration accuracy.

AUG automatic image registration/multi-layer registration technology is based on an area-based feature-matching algorithm that intelligently identifies control points between two images with overlapping fields of view. A novel consistency check determines the number of correctly identified control points and removes mismatched points between the images. It then aligns one image with the other using different transformation functions including linear conformal, affine, projective, and high order polynomial transformation. The registered images can achieve user defined sub-pixel accuracy in Root-Mean-Square (RMS) error sense as a metric in measuring registration accuracy.

A recent breakthrough has allowed AUG Signals to take our advanced registration capabilities to a new level. The uniqueness of the new image/vector conflation and electro-optical (EO)/synthetic aperture radar (SAR) registration technology lies in its simultaneous utilization of multiple common feature image pairs generated from originally different data sources. The technology produces unparalleled in their accuracy results, overcoming the inherent difficulties of automatic registration of different image characteristics.


  • Fully automatic multi-layer co-registration of images from similar or dissimilar sensors
    • Increased number of automatically identified control points between multi-sensor image pair;
    • Increased spatial registration accuracy for multi-sensor image co-registration;
    • Increased similarities between the images by extracting feature layers;/li>
    • Automatic vector-to-image data conflation
  • Sub-pixel accuracy – in the RMS error sense
  • Automatic registration capabilities that employ different transformation functions,
  • Intelligent and efficient control points estimation and global consistency checking to eliminate mismatched points
  • Robust and works for many different kinds of multi-sensor images




Original LandSAT 7 image of first three bands as RGB image, Ottawa, July 26, 2007


Original RADARSAT1 image, Ottawa region, July 17, 2007


Road vector file, Ottawa region, July 17, 2007


A spatial subset of registered LandSAT overlaid with registered Road vector file to show the accuracy of the registration result.
The roads from the vector file are perfectly aligned with the road in the LandSAT image.
Control points: 54, RMS error: 0.7151


A spatial subset of registered RADARSAT overlaid with registered Road vector file.
The roads from the vector file are perfectly aligned with the road in the RADARSAT image.
Control points: 35, RMS error: 0.6324


A spatial subset of registered LandSAT overlaid with registered RADARSAT.
The two images are perfectly aligned.
Control points: 42, RMS error: 0.5271

Accurate registration opens the door to a number of other automated techniques that previously have been carried out primarily using analyst intensive techniques: change detection and target detection and identification.