FINE

In this chapter the processing of step FINE is described. The offset vectors to align the slave image to the master are computed with sub pixel accuracy for a number of locations in the master. Over the total image, for a large number of windows (e.g., 500, distributed by Doris or from a file with locations in the master coordinate system), the offset between master and slave is estimated by computing the correlation of the magnitude images for shifts at pixel level. Next, in a local neighborhood of the maximum (correlation at pixel level) these correlations are harmonically oversampled (interpolated, requires FFT) to find the maximum at sub pixel level. These offsets are then written to the (products) result file. The offset is computed in the spectral or in the space domain (which is implemented to avoid the use of FFT, but that is required later anyway, and to provide a check of the method in the spectral domain, which should be faster). The correlation is computed on the magnitude images. Though we believe this to be a good method, we would like to investigate first oversampling the images itself, and directly computing the correlation for a small number of shifts (assuming initial offsets are known within a few pixels), as we suspect that there may be an error introduced due to aliasing with the method that is implemented. (This method will be named 'oversample'.)

The actual computation of the transformation model (2d polynomial) is done by the step COREGPM (computation of coregistration parameters). See also [13].

Figure 17.1: Plot produced by the command 'plotoffsets interferogram.out 11 6000 21 1000 0.6 Outdata/1393.raw' (keycard FC_PLOT 0.6 BG). The magnitude is plotted in the background. Correlation is indicated by the size of the circles, estimates with a correlation below 0.6 are filtered out.