The ASTER instrument configuration with the multi-telescopes necessitates routine processing for the inter-telescope band-to-band registration on the ground, unless the boresights are stable enough during the mission life to keep the initial state after launch. The pointing change mechanism is also a source of misregistration because of limited accuracy of pointing position knowledge.

The important feature to be stressed here for the inter-telescope registration is that a set of correction parameters will be valid and can be applied to all images as long as the pointing is kept at the same position and the elapsed time since a previous parameter setting is short enough for boresight stability. The inter-telescope registration correction will be carried out every observation unit, using reference bands of each telescope (band 2 for VNIR, band 6 for SWIR and band 11 for TIR).

There is no strong reason for this reference band selection, although a matching experiment was carried out using simulated image data. Band 2 was slightly better as a reference band than other VNIR bands because of a relatively large atmospheric absorption in band 1 and a peculiarity of band 3 for vegetation. Band 6 is selected as reference band of SWIR because it is one of the bands used for parallax correction. Band 11 is selected as reference band of TIR without any major reason.

Figure 2-9 shows the inter-telescope geometric correction flows for VNIR band 2. This correction process is carried out as follows.

1. Start the registration error calculation at the beginning of each observation unit.
2. Select band 2 as a moving window image and band 6 or 11 as a target image. The moving window is selected such that its center corresponds to the lattice point at which the geometric correction is carried out. The target image is selected to cover the search area. The moving window image size is 41 x 41 pixels. The target window size is larger than the moving window to cover the search area.
3. Select a cloud free window for the correlation by repeating the previous item 2.
4. Carry out the radiometric correction for the cloud free images.
5. Find the correlation coefficients by moving the moving window in pixel units in both along-track and the cross-track directions.
6. Find the point of highest correlation in sub-pixel units by interpolating the correlation data calculation in the pixel unit.
7. Evaluate the image matching quality. The criteria for judgment is the correlation coefficient. The threshold value for the correlation coefficient is set to 0.7 in the initial stage.
8. Repeat the process from item (2) to item (7) a large number of times (between 100 and 200), selecting other target images, to enhance the accuracy by averaging.
9. If the preset number of error data can not be reached in an observation unit, the inter-telescope registration correction will not be carried out. The failed information will be reported.
10. Exclude the error data which deviates over 3s value from the average.
11. The obtained number of the effective error data is averaged to generate a set of final error data.
12. Calculate 3s value to evaluate the accuracy.

An evaluation of the algorithm was carried out using simulated images generated from Landsat TM and airborne sensor images. The registration accuracies in 3s for SWIR/VNIR were 0.054 SWIR pixels and 0.051 SWIR pixels in the cross-track and the along-track directions, respectively. The registration accuracies in 3s for TIR/VNIR were 0.050 TIR pixels and 0.044 TIR pixels in the cross-track and the along-track directions, respectively. Judging from these results a required accuracy of 0.3 pixels (3s) for the inter-telescope registration will be achievable by averaging the image matching data in the same observation unit, if the boresight of each telescope is stable during a maximum observation time of 16 minutes. A hundred data points will be enough for averaging to satisfy a required accuracy of 0.3 pixels in 3s.