Comparative Evaluation of the Influence of the Terrain Topography on the Registeration Accuracy of the Fused Multi- Sensor Satellite Images (case Study: P5 and P6 Irs Satellite Images)

Multi sensor image fusions, by which complementary information contents of the acquired images are combined, make image analysis easier for variety of applications. However, any subsequent application carried out on the fused images highly relies on the image fusion accuracy. In this regard, the image registration stage plays a crucial impact on the final accuracy. Inaccurate image registration disturbs the positional correspondences between the identical pixels and hence introduces uncertainties on the subsequent data analysis carried out over the integrated data. The common approach for the image to image registration employs a polynomial method using well distributed image points. However, for the images taken by different sensors, simple polynomial transformation for the registration may not give satisfactory result in the areas with excessive height undulation. This does not pose a real problem for the low resolution satellite images for the reason that the relief displacements in these images are more or less negligible. Nevertheless, for the high resolution satellite images, the relief displacement may not be negligible particularly in mountainous terrains where it may reach to several pixels. In these situations the only solution is to incorporate a DTM of the area to correct the relief displacement. However, polynomial transformation is not an optimal solution to perform this task. In this paper, an alternative approach for the image registration is proposed based on a joint RPC/3D-Affine transformation. The proposed solution is applied to the IRS P5 and P6 satellite images. The main characteristic of this approach is that instead of performing a direct image to image registration, the geometric transformation is carried out first from the image to ground via the P5 supplied RPCs. This is then followed by a 3D affine transformation from the ground to the P6 image space. The second stage requires the parameters of the 3D affine transformation. These are determined in a preliminary stage. This joint approach allows the elimination of the relief displacement and hence can register the P5 and P6 images with a pixel accuracy level. In this paper first the insufficiency of the image to image transformation is presented by evaluation of the polynomial methods performed on the images taken over the flat and mountainous terrains. The suitability of the proposed RPC/3D-Affine transformation method for the fusion of the P5/P6 images is then demonstrated.