The large offset among SWIR bands in the along-track direction shown in Figure 2-8(a) gives rise to parallax errors for band-to-band registration depending on the distance between the instrument and the targeted ground. The algorithm used for SWIR intra-telescope registration is a combination of the image matching correlation method and the coarse DEM method. The ETOPO-030 prepared by EROS Data Center will be used as the coarse DEM.

A criterion of selecting band for image matching is good correlation, consequently spectrally close. Bands 6 and 7 are selected, although any other combination of neighbouring bands in the 2µm region will be available for this purpose. As shown in Figure 2-8(a), detector arrays of bands 6 and 7 are located at opposite ends of the focal plane to increase the parallax detection sensitivity.

Figure 2-8(b) shows the parallax correction flow. In order to eliminate this band-to-band misregistration due to the parallax, the SWIR parallax correction process handles the image data as follows.

1. Select band 7 as a moving window image and band 6 as a target image. The moving window is selected such that its center corresponds to the lattice point at which the geometric system correction is carried out. The target image is selected to cover the search area. The moving window image (band 7) size is 21 by 21 pixel.
2. Select only cloud free windows for the correlation. Use the coarse DEM data for cloudy windows.
3. Carry out the radiometric correction for the two window images.
4. Find correlation coefficients by moving the window in the along-track direction.
5. Find the highest correlation point in sub-pixel units by interpolating the correlation data calculated in pixel units.
6. Evaluate the image matching quality. Criteria for judgment are the correlation coefficient and the deviation from the predetermined value from the coarse DEM data. The threshold value for the correlation coefficient and the threshold deviation for the predetermined value are set to 0.7 and 20% deviation, respectively, in the initial stage. Use the parallax errors calculated by the coarse DEM if the image matching quality can not meet the criterion mentioned above.
7. Evaluate the parallax error of all SWIR bands for the nadir direction.

The registration error due to the parallax is evaluated at every lattice point (the corner point of the block) and expressed with the pixel unit for the Level-0B image data in the along-track direction.

An evaluation of the algorithm was carried out using simulation images generated from airborne sensors and Landsat TM images. Miss-registration due to parallax error was intentionally introduced to these images using DEM data. Judging from these results 99 % of the data are within 0.3 pixels for all images.