Tukukkk

The Daily Galaxy article “The Lost Hubble” –New! Deepest Image of the Universe Ever Taken says:

To produce the image a group of researchers from the Instituto de Astrofísica de Canarias (IAC) led by Alejandro S. Borlaff used original HUDF images from the Hubble Space Telescope. After improving the process of combining several images the group was able to recover a large quantity of light from the outer zones of the largest galaxies in the HUDF. Recovering this light, emitted by the stars in these outer zones, was equivalent to recovering the light from a complete galaxy (“smeared out” over the whole field) and for some galaxies this missing light shows that they have diameters almost twice as big as previously measured.

The image looks really strange, what is going on? Is there a technical article associated with this work?

When you plug the lead researcher's name into Arxiv, the first search result is The missing light of the Hubble Ultra Deep Field.

3 main steps:

• Creation of sky flat fields for the four filters. This process is fully described in Sect. 2.4.

– Creation of a catalogue of all WFC3/IR datasets that may affect our mosaics (including calibration exposures) to generate a set of improved persistence models for each exposure of the HUDF. We detail this process in Sect. 2.5.

– Download and reduction of all the WFC3/IR datasets that include observations using the F105W, F125W, F140W and F160W filters on the HUDF.

Flat sky field:

In order to measure the relative sensitivity of the pixels of a detector (flat field), the optimal process would be to observe a uniform external source of light.

Basically they're trying to remove all sources of noise from the image, in an attempt to make faint signals appear in places where that signal has been overwhelmed by noise.

Persistence models:

A known effect that affects HgCdTe IR array detectors (as is the case of the WFC3/IR) is persistence. Persistence shows up as an afterglow on the pixels that were exposed to a bright source of light in a previous exposure.

The current method of persistence correction of WFC3/IR con- sists in modeling the number of electrons that would be created by persistence in each pixel by all the previous exposures (up to a certain time) that were taken before the one to correct (Long et al. 2012).

During long exposures, sky background can vary noticeably, introducing a non-linear component to the count rates calculated by calwf3.

We individually estimate and subtract the sky background emission from each readout of the intermediate ima.fits files.

In order to avoid systematic biases due to the presence of de- fects in some regions of the detector, we created a manual data quality mask to flag those regions were the flat field cannot fully correct the differences in sensitivity.

More image processing to remove sky background:

In this Section we describe the methods used to remove the sky background from the individual exposures and the final mosaics of the HUDF.

Image alignment:

As a consequence, when comparing images from different visits, it is usual to see that they are not exactly aligned. In order to exploit the full resolving capabilities of WFC3, we need to carefully re-align the images of different visits to a single reference world coordinate system solution (WCS hereafter).

and as a final step, image combination.

Result:

The XDF version of the HUDF WFC3/IR mosaics is dominated by a systematic bias in the form of a significant oversubtraction of the sky background around the objects with large angular size. A similar result (to a lesser extent) is obtained for the HUDF12. We successfully recover a significant amount of over-subtracted diffuse light around the largest objects of the HUDF, not detected by the previous versions of the mosaics.

Summary:

They've processed the images to bring out details in the galaxies. In the space between the galaxies, the image processing gives garbage results (the white areas), but they've managed to bring out detail on the edge of the galaxies that was hidden before.

In response to a couple comments that Hobbes' answer is a bit thick, how about:

To reduce noise effects, the team did flat-fielding adjustment and then summed multiple exposures, thus allowing weak signals to add while noise effects cancelled out.

That's the TL;DR which leaves out a lot of really cool methods of identifying "true dark" and noise patches vs. reliable signals (stars or galaxies or whatever).