Updating geographic information using high resolution satellite images has become a major competitor to the traditional photogrammetric works. This research presents a new technique to achieve geometric correction, starting with automatic satellite imagery matching with digital photogrammetric data, after outliers' exclusion. Matched points are ortho-corrected using DDTM. A downward Multi-layer perceptron neural networks technique will be used in the process of network training, instead of using the classic upward technique. In the new training process image coordinates were used as inputs and their corresponding ground coordinates were used as outputs. The trained network was used in predicting ground coordinates of a set of new regularized image points in the same space domain of the matched point dataset. Rational function model (RFM) will be implemented using regularized ortho-corrected points as GCPs in order to reach the final relationship parameters between satellite imagery and the 3D object coordinates. The new technique led to an improvement of the accuracy by damping down the error to 0.67 the error resulting from the conventional RFM model. |
