The idea of working on GALEX color images arose when preparing an Image cutout service for GALEX, similar to the one already existing for SDSS data. We started using the jpeg images from the pipeline; to be used within a cutout service, we needed at least to remove the annotations, be able to mask out to apply the primary resolution etc ... Additional requirements for these images are to present an homogeneous background; we wanted also to be able to distinguish bright to faint objects, and finally, display false colors representative of the range of FUV-NUV colors present in the data. We tried to build color images fullfilling these conditions.
We began working directly with the RGB color scheme. A basic color mapping consists to set the R channel as NUV, the B channel as FUV, and the G channel as a combination of FUV and NUV. It has been however quite difficult to get a wide range of colors in this scheme; this partly because we tried to map two independent informations on three channels. We tried to use for the G channel a non linear function of the FUV over NUV ratio, with no significant results. We then decided to give up the RGB color space to the HSV (Hue Saturation Value). In this color space, Hue decribes the dominant color, and is described by a number between 0 (meaning red) and 360 (purple). The saturation (0 < s < 1) sets the dullness of a given color, s = 0 coding the dullest. Finally the value v (0 < v < 1) sets the brightness.
More specifically, we first determine the mean and standard deviation
of the fluxes in the FUV and the NUV images. To do so, we slightly
filter the images, using a gaussian filter of 1.5 fwhm. This is
required as the actual fluxes show quanticized values, due to the
nature of the detector. These values enable to build a
chi2 image defined as follow: