The IPC step corrects a JWST exposure for interpixel capacitance by convolving with an IPC reference image.
The current implementation uses an IPC reference file that is normally a small, rectangular image (e.g. 3 x 3 pixels), a deconvolution kernel. The kernel may, however, be a 4-D array (e.g. 3 x 3 x 2048 x 2048), to allow the IPC correction to vary across the detector.
For each integration in the input science data, the data are corrected group-by-group by convolving with the kernel. Reference pixels are not included in the convolution; that is, their values will not be changed, and when the kernel overlaps a region of reference pixels, those pixels contribute a value of zero to the convolution. The ERR and DQ arrays will not be modified.
Subarrays are treated the same as full-frame data, with the exception that the reference pixels may be absent.