CROSS-CORR employs a pattern-matching technique which depends heavily upon the linear cross-correlation function to identify similar patterns between a raw science image and the raw-space ITF. A rectangular grid of fiducial positions (centered 28 pixels apart) has been defined for each camera and dispersion. At each defined position, a cross correlation is performed to determine the displacement between the raw image and the ITF at that locale. Specifically, at each locale an array (template) of 23 × 23 pixels of the raw science image is used for comparison to a corresponding array (window) of 29 × 29 pixels of the raw-space ITF image. The window is larger than the template in both dimensions such that the template can be moved about the window to ``search'' for the best correlation location. It is known from a number of analyses that the maximum extent to the motion of the fixed pattern in most raw images is only a few pixels. Therefore, the window size is ordinarily set in the NEWSIPS processing to accommodate motion of only a few pixels, but the size can be varied, if necessary.
Once a pattern match location is found to whole pixel accuracy at a defined fiducial locale, the match location is evaluated via a series of statistical tests to validate the result. Upon validation, further computations refine this determination to the sub-pixel accuracy required for an accurate photometric correction.
After all the valid pattern match locations for a camera and dispersion have been derived to sub-pixel accuracy for the raw science image, the determinations are viewed as spatial deviations and examined as an ensemble for consistency. Displacements which do not conform in a general sense to a smoothly varying function are adjusted to conform. Finally, after all the fine tuning has been completed, a full three-dimensional (3-D) displacement image ( 768 × 768 × 2)is created for Sample and Line using a bi-cubic spline interpolation scheme.