CROSS does a cross-correlation on two images, loading the resulting image into a third buffer. This program is useful for registering two images or determining a shift between two spectra, for instance. It is usual practice to subtract an appropriate baseline from the images, usually the sky level for images and the continuum level for spectra. The RAD=r keyword defines the half-size of the cross-correlation image. If not specified the cross-correlation will be the same size as the first source image. However, a cross-correlation of a two-D image involves four nested DO loops, so it is advisable to use the RAD= keyword when possible so that two of the DO loops are short. Often only the peak of the cross-correlation is of interest, so that the a very small radius may be specified (between 2 and 5, say). Spectra are automatically cross-correlated in one-dimension only.
The CORR keyword makes the routine compute full correlation coefficients rather than just cross-correlations as a function of lag. This option involved about a factor of 3 more operations, so runs slower. It is important for radial velocities or displacements because the correlation peak tends to be more symmetric than the cross-correlation peak. Asymmetries in cross-correlation and in correlation arise because both are computed over a finite region.
The peak of the cross-correlation image may be found using the SURFACE
command with the LOAD keyword as in the following example, which
cross-correlates two images, calculates the peak of the cross-correlation,
and then shifts the two images into register: