Improved Non-Parametric Geometric Corrections For Satellite Imagery Through Covariance Constraints

Satellite imagery enables rapid data collection and assessment for several earth and environmental sciences. However, the obtained raw imagery cannot be used directly for spatial data collection and assessment and should be further processed for geometric distortions. The usefulness and reliability of the data obtained from satellite images heavily relies on the geometric correction process and it can be applied through either parametric or non-parametric methods. Parametric methods require physical sensor model parameters which are generally available only to the vendor. The Rational Polynomial Coefficients (RPCs) supplied by several vendors provide global geometric correction, however they are usually available for specific high resolution imagery systems. Other polynomials methods provide geometric corrections locally but applicable for nearly all satellite imagery systems. Depending on the degree of the employed polynomial, the topography, the distribution and the quality of the control points, local geometric corrections could provide lower misfits at GCPs (Ground Control Points) while they might produce very large misfits at ICP (Independent Check Points). In particular, when the available number of GCPs is limited and they are poorly distributed over the image area. In this study, 2D and 3D Affine functions with covariance constraints are introduced to improve the geometric correction when the available GCP are of limited accuracy and poorly distributed. The efficiency of the proposed method was numerically shown in two applications. The results show that the proposed method provides a compromise between the local and global misfits without using any ICP coordinates and enables robust and efficient geometric correction of satellite imagery which lacks a precise RPC model such as TUBİTAK RASAT satellite.