After having been fed through the pipeline, the registration of the individual images was checked by comparing the positions of several bright sources. This revealed a couple of unexpected large (0.8 arcsec) shifts, but otherwise suggested that the images at each dither position were registered well to within the errors of such a comparison. The images with the large shifts have been treated as separate dither positions. Subsequent tests have revealed small (~0.1 pixel) shifts between some frames that were nominally at the same dither position. These positional errors have not yet been corrected; however, their influence on the final image is expected to be negligible.
The roughly 5 images per dither position were cosmic-ray rejected and stacked into a single image. Shifts and rotations between these dither stacks were measured using a specially written cross-correlation task. Estimated errors are a few hundredths of a pixel for F450W, F606W, and F814W. Only two dither positions -- those produced by the unexpected shifts -- showed any measurable rotation.
Once the shifts and rotations were measured, the images were aligned and combined using the drizzle algorithm, which corrects for geometric distortion, and produces an output image that is sampled on a smaller pixel scale than the input images. The final pixel scale is 0.04 arcsec for both the WF and the PC chips.
Copyright © 1997 The Association of Universities for Research in Astronomy, Inc. All Rights Reserved.