Comparison of Different Mathematical Models on the Accuracy of the Orthorectification of Aster Imagery

The accuracy assessments of the orthorectification of ASTER imagery carried out using twelve different mathematical models are presented. For performing the orthorectification, 77 ground control points (GCPs) and a digital elevation model (DEM) generated from stereo ASTER images were utilized. The mathematical models used include two rigorous and ten simple geometric models. Among the twelve models, the Toutin’s Rigorous Model (Toutin, 2002) was the only model implemented using a commercial image analysis software. The mathematical models for the remaining eleven models were developed using Matlab 6.5.0. In order to find the effect of the number of GCPs on the accuracy of orthorectification, the ASTER image was rectified using evenly distributed 40, 50, 60, 70 and 77 GCPs. It was found that, the models First Order 2D Polynomial, First Order Polynomial with Relief and Direct Linear Transformation provided the highest RMS errors of 2.41, 1.81, and 1.62 pixels, respectively. The RMS errors computed for the models Projective Transformation, Second Order 2D Polynomial, and Third Order 2D Polynomial were 1.16, 1.15, and 0.99 pixels, respectively. The accuracies of the Orun and Natarajan Model (Orun and Natarajan, 1994) and Toutin’s Rigorous Model were found to be 0.75 pixels. Likewise, the model First Order Rational Functions produced an accuracy of 0.73 pixels. The models Second Order Rational Functions, Third Order Rational Functions, and the Second Order Polynomial with Relief, which was developed in this study, provided accuracies in the order of half a pixel. It was found that the model Second Order Rational Functions provided the best results for the orthorectification of the ASTER images. However, the developed model Second Order Polynomial with Relief provides simplicity, consistency, and requires less number of GCPs when compared to the model Second Order Rational Function.