The new processing algorithms that have been developed by the NASA IUE Project allow several significant improvements in the processed data. The new approach exploits the presence of fixed pattern noise (pixel-to-pixel sensitivity variations in the cameras) as a reliable fiducial to register the raw science image with the raw Intensity Transfer Function (ITF) image. Proper registration of IUE images is crucial to accurate photometric correction because the variability of the geometrical distortions introduced by the SEC-Vidicon cameras ensures that raw science images are never perfectly aligned with the ITF. While reseau marks etched on the faceplates of the cameras were intended to be used to rectify geometrically the science images, they cannot be detected at the low exposure levels usually found in the background of IUE images. Therefore, the IUESIPS method of processing IUE images uses predicted reseau positions to align the science images with the ITF images. Unfortunately, these mean positions are poorly known and the application of a mis-registered ITF (by more than about 0.2 pixel) manifests itself as systematic noise in the photometrically corrected image, and ultimately in the spectrum.
To achieve proper alignment of the ITF images with each science image for the Final Archive reprocessing, the fixed pattern inherent in IUE images is used as a fiducial. Small patches of the science image are cross-correlated against corresponding areas on the appropriate ITF image to determine the spatial displacement between these two images. The displacement of each pixel in the science image from its corresponding pixel in the ITF can thus be determined to sub-pixel accuracy. Such an approach has several advantages: (1) a large number of fiducials can be found anywhere on the image, (2) fixed pattern can be detected even at the lowest exposure levels, and (3) fiducials are available near the edge of the image, where distortion is greatest. In the IUESIPS processing of IUE data, the ITF images have been resampled to geometrically correct space, significantly smoothing these calibration data. In the new processing system, the ITF images are retained in raw space, increasing the accuracy of the pixel-to-pixel photometric correction.
Only one resampling of the data is performed in the new processing system, minimizing the smoothing inherent in such an operation. The linearized pixel values are resampled into a geometrically rectified and rotated image, such that the spectral orders are horizontal in the image and the dispersion function of the spectral data within an order is linearized. The resampling algorithm used is a modified Shepard method which preserves not only the flux to 1-3in the image, but also the spectral line shapes.
The low-dispersion spectral data are extracted by a weighted slit extraction method developed by Kinney et al. (1991). The advantages of this method over the IUESIPS boxcar extraction are: (1) the signal-to-noise ratio (S/N) of the spectrum is usually improved while flux is conserved, (2) most of the cosmic rays are automatically removed, and (3) the output includes an error estimate for each point in the flux spectrum. The high-dispersion spectral data are extracted using an IUESIPS style boxcar extraction method. As a result the S/N improvements may not be as good as those seen in low-dispersion data.
An entirely new data product for the IUE Final Archive is a geometrically rectified and rotated high-dispersion image, with horizontal spectral orders. This new data product will allow future investigators to perform customized extractions and background determinations on the high-dispersion data. One of the most significant problems with the analysis of high-dispersion IUE data has been the proper determination of the background in the region where the echelle orders are most closely spaced and begin to overlap. The new processing system includes a background removal algorithm that determines the background level of each high-dispersion image by fitting, in succession, one-dimensional Chebyshev polynomials, first in the spatial and then the wavelength direction. The extracted high-dispersion spectral data are available order-by-order with wavelengths uniformly sampled within an order.
In addition to the new algorithms for processing the IUE data for the
Final Archive, all absolute flux calibrations have been rederived. The
new calibrations use white dwarf models to determine the relative shapes
of the instrumental sensitivity functions, while previous UV satellite
and rocket observations of 
UMa and other standard stars are used
to set the overall flux scale. The IUE Final Archive extracted
spectral data are also corrected for sensitivity degradation of the
detectors over time and temperature, a calibration not previously
available with IUESIPS processing.
These new processing algorithms for the creation of the Final Archive allow a significant improvement in the signal-to-noise ratio of the processed data, resulting largely from a more accurate photometric correction of the fluxes and weighted slit extraction, and greater spectral resolution due to a more accurate resampling of the data. Improvement in the signal-to-noise ratio of the extracted low-dispersion spectral data has been shown to range from 10-50% for most images, with factors of 2-4 improvement in cases of high-background and underexposed data (Nichols-Bohlin 1990).