---

7.2.1 NoiseChisel changes after publication ¶

NoiseChisel was initially introduced in Akhlaghi and Ichikawa 2015 and updates after the first four years were published in Akhlaghi 2019. To help in understanding how it works, those papers have many figures showing every step on multiple mock and real examples. We recommended to read these papers for a good understanding of what it does and how each parameter influences the output.

However, the papers cannot be updated anymore, but NoiseChisel has evolved (and will continue to do so): better algorithms or steps have been found and implemented and some options have been added, removed or changed behavior. This book is thus the final and definitive guide to NoiseChisel. The aim of this section is to make the transition from the papers above to the installed version on your system, as smooth as possible with the list below. For a more detailed list of changes in each Gnuastro version, please see the NEWS file²⁰⁴.

• An improved outlier rejection for identifying tiles without any signal has been implemented in the quantile-threshold phase: Prior to version 0.14, outliers were defined globally: the distribution of all tiles with an acceptable --meanmedqdiff was inspected and outliers were found and rejected. However, this caused problems when there are strong gradients over the image (for example, an image prior to flat-fielding, or in the presence of a large foreground galaxy). In these cases, the faint wings of galaxies/stars could be mistakenly identified as Sky (leaving a footprint of the object on the Sky output) and wrongly subtracted.

  It was possible to play with the parameters to correct this for that particular dataset, but that was frustrating. Therefore from version 0.14, instead of finding outliers from the full tile distribution, we now measure the slope of the tile’s nearby tiles and find outliers locally. Three options have been added to configure this part of NoiseChisel: --outliernumngb, --outliersclip and --outliersigma. For more on the local outlier-by-distance algorithm and the definition of slope mentioned above, see Quantifying signal in a tile. In our tests, this gave a much improved estimate of the quantile thresholds and final Sky values with default values.