When datasets are offset in their Current attached coordinate system only by X and Y translations as above, but contain insufficient objects for the fully automated registration described in the previous section, registration may be done in a semi-automated way, using the programs:
PAIRNDF displays the images to be registered and allows the direct selection of image features which are common between image pairs. This works by asking you alternately to choose a pair of images which has an overlap, and then to line them up. When enough of the datasets have been selected this way, and successfully paired, the global correspondence of the image features is determined. Using this method avoids the need to identify each image feature in a particular sequence, they are just selected as being in `common' between any pair, PAIRNDF works out the rest using the same methods as FINDOFF, except a single spanning tree is not chosen since in this case, all pairings are assumed to be correct.
Both the pair selection and the pair alignment are done using an intuitive graphical user interface on the Xwindows display. During pair selection you can see any combination of two images side by side, along with their names and other information such as selected FITS headers. They are displayed resampled into their current coordinate frames, so that it should be easy to see where any overlap between the two is. When you have selected a pair which contains objects common to both, you can move to the alignment stage in which you drag one image with the mouse to the correct position on top of the other. The display can be zoomed and scrolled to make careful positioning easy. You are then asked to select centroidable features in the overlap region so that the program can determine the offset between the two accurately.
To summarise, during the graphical part of PAIRNDF you have to do the following:
As with all graphically based interfaces it is difficult to describe the advantages and ease of use, it's best to try them out.
CCDPACK