Image-to-Image Registration Quality Assessment of IKONOS/QuickBird Imagery Using Aerial Orthoimagery

An Image-to-Image registration assessment was conducted on IKONOS Precision Master imagery and QuickBird Standard imagery of Sioux Falls, South Dakota. Independent confirmation of vendor accuracy specifications is essential for instilling customer confidence in the product for potential applications. In our case, the application in mind is providing ground control and reference imagery for geometric characterization and calibration of Landsats 5 and 7, as well as future satellite missions. In particular, we are interested in obtaining international ground control points (GCP) at least as accurate as the GCPs in our CONUS library, which are predominantly DOQ control points accurate to ~12-meters CE90 or better. Availability of control with that level of accuracy has been difficult to obtain overseas. Therefore, the primary goal of the study was to measure registration quality of the vendor’s products against an orthophoto mosaic, acquired by a local aerial surveying company, and having a horizontal accuracy of ~1 ft. (0.3 meters) at its original 0.5 ft. resolution. The imagery was purchased by the city of Sioux Falls for their engineering GIS and provided to us for the purpose of this study. The Landsat 7 Image Assessment System (IAS) image-to-image characterization software was used to measure image offsets and compute error statistics. The software was developed by the USGS, and although general-purpose in nature, its traditional role is image-to-image characterization of Landsat 7’s 15-meter panchromatic band. Thus, a secondary goal of the project was to evaluate the viability of using the IAS to characterize the higher spatial resolutions of the IKONOS and QuickBird panchromatic products (1-meter and 0.6-meter resolution, respectively). The IAS software uses an area-based matching algorithm based on normalized cross-correlation and correlation surface-fitting to obtain the sub-pixel location of the peak. Requirements dictate the software to be capable of correlating common features in the same bands of separate Landsat 7 images to an accuracy of 0.1 pixel at the 90% probability level. To verify the results based on this automated technique, tie-points were also selected manually and statistics based on the IAS automated matching method were compared with statistics based on the manual method.