Comparative Analysis of Area-based Image Matching Techniques for High Resolution Satellite Imagery

Recently, it has become much easier to obtain commercial high-resolution satellite imagery (HRSI), and application of IKONOS satellite imagery to 3D positioning has recently been investigated, with sub-meter accuracy being achieved. Image matching is one of the most fundamental processes in geopositioning from multi-image HRSI. Least squares matching and cross correlation matching are the most basic and popular techniques applied by photogrammetrists. These techniques, however, have most often been used independently. This paper presents the initial results of a comparative analysis of these two image matching techniques. Initially, the relationship between the two techniques is described. Maximising the cross correlation coefficient is equal to minimising the sum of squared distances between observed points and a regression line in feature space. This interpretation leads to principal component regression. By using principal components, the cross correlation coefficient can be maximised in the same way as in least-squares estimation. Accordingly, we develop a cross correlation matching which includes not only translation but also deformation (affine and projective transformation). The least squares matching and cross correlation matching are applied to an IKONOS stereo pair, and the characteristics and results of applying the matching techniques are discussed.