Assessing Geometric Reliability of Corrected Images from Very High Resolution Satellites

Since the launch of Ikonos by Space Imaging, LLC on 24 September 1999, the very high resolution (VHR) satellite imagery has been applied to diverse fields. Every application needs a certain geometric accuracy in the corrected image; therefore, the planimetric accuracy control of VHR satellite imagery proves to be fundamental. As a rule of thumb, the Root Mean Square error (RMS) computed at independent check points (ICPs) is the global measure most widely used for accuracy assessment in VHR imagery. This paper presents an assessment, focused on two QuickBird and Ikonos panchromatic single images, of the number of ICPs required to obtain an estimation of one-dimensional accuracy (RMS1d) with a certain confidence level or reliability. Thus, two theoretical approaches have been tested to estimate reliability depending on the number of ICPs, and they have been experimentally validated using the Monte Carlo simulation method. The residual’s samples were generated for both satellite images in the best possible operational conditions: (a) using optimal sensor models, (b) with high accuracy ground points measured by Differential Global Positioning System, (c) with an adequate number of well distributed ground control points (GCPs), and (d) using GCPs and ICPs well-defined on the raw images, i.e., with a reasonably low pointing error. Under these conditions, the two theoretical models tested provided a good fit (r 2 �97 percent) for the simulated data offered by Monte Carlo when outliers were withdrawn. There were no notable differences between the results obtained from the Ikonos and QuickBird scenes.