GNU Astronomy Utilities



2.1.10 NoiseChisel and Multi-Extension FITS files

In the previous sections, we completed a review of the basics of Gnuastro’s programs. We are now ready to do some more serious analysis on the downloaded images: extract the pixels containing signal from the image, find sub-structure of the extracted signal, do measurements over the extracted objects and analyze them (finding certain objects of interest in the image).

The first step is to separate the signal (galaxies or stars) from the background noise in the image. We will be using the results of Dataset inspection and cropping, so be sure you already have them. Gnuastro has NoiseChisel for this job. But NoiseChisel’s output is a multi-extension FITS file, therefore to better understand how to use NoiseChisel, let’s take a look at multi-extension FITS files and how you can interact with them.

In the FITS format, each extension contains a separate dataset (image in this case). You can get basic information about the extensions in a FITS file with Gnuastro’s Fits program (see Fits). To start with, let’s run NoiseChisel without any options, then use Gnuastro’s Fits program to inspect the number of extensions in this file.

$ astnoisechisel flat-ir/xdf-f160w.fits
$ astfits xdf-f160w_detected.fits

From the output list, we see that NoiseChisel’s output contains 5 extensions. The zero-th (counting from zero, with name NOISECHISEL-CONFIG) is empty: it has value of 0 in the fourth column (which shows its size in pixels). Like NoiseChisel, in all of Gnuastro’s programs, the first (or zero-th) extension of the output only contains meta-data: data about/describing the datasets within (all) the output’s extensions. This is recommended by the FITS standard, see Fits for more. In the case of Gnuastro’s programs, this generic zero-th/meta-data extension (for the whole file) contains all the configuration options of the program that created the file.

Metadata regarding how the analysis was done (or a dataset was created) is very important for higher-level analysis and reproducibility. Therefore, Let’s first take a closer look at the NOISECHISEL-CONFIG extension. If you specify a special header in the FITS file, Gnuastro’s Fits program will print the header keywords (metadata) of that extension. You can either specify the HDU/extension counter (starting from 0), or name. Therefore, the two commands below are identical for this file. We are usually tempted to use the first (shorter format), but when putting your commands into a script, please use the second format which is more human-friendly and understandable for readers of your code who may not know what is in the 0-th extension (this includes yourself in a few months!):

$ astfits xdf-f160w_detected.fits -h0
$ astfits xdf-f160w_detected.fits -hNOISECHISEL-CONFIG

The first group of FITS header keywords you see (containing the SIMPLE and BITPIX keywords; before the first empty line) are standard keywords. They are required by the FITS standard and must be present in any FITS extension. The second group starts with the input file name (value to the INPUT keyword). The rest of the keywords you see afterwards have the same name as NoiseChisel’s options, and the value used by NoiseChisel in this run is shown after the = sign. Finally, the last group (starting with DATE) contains the date and version information of Gnuastro and its dependencies that were used to generate this file. Besides the option values, these are also critical for future reproducibility of the result (you may update Gnuastro or its dependencies, and they may behave differently afterwards). The “versions and date” group of keywords are present in all Gnuastro’s FITS extension outputs, for more see Output FITS files.

Note that if a keyword name is larger than 8 characters, it is preceded by a HIERARCH keyword and that all keyword names are in capital letters. These are all part of the FITS standard and originate from its history. But in short, both can be ignored! For example, with the commands below, let’s see at first what the default values are, and then just check the value of --detgrowquant option (using the -P option described in Option management and configuration files).

$ astnoisechisle -P
$ astnoisechisel -P | grep detgrowquant

To confirm that NoiseChisel used this value when we ran it above, let’s use grep to extract the keyword line with detgrowquant from the metadata extension. However, as you saw above, keyword names in the header is in all caps. So we need to ask grep to ignore case with the -i option.

$ astfits xdf-f160w_detected.fits -h0 | grep -i detgrowquant

In the output of the above command, you see HIERARCH at the start of the line. According to the FITS standard, HIERARCH is placed at the start of all keywords that have a name that is more than 8 characters long. Both the all-caps and the HIERARCH keyword can be annoying when you want to read/check the value. Therefore, the best solution is to use the --keyvalue option of Gnuastro’s astfits program as shown below. With it, you do not have to worry about HIERARCH or the case of the name (FITS keyword names are not case-sensitive).

$ astfits xdf-f160w_detected.fits -h0 --keyvalue=detgrowquant -q

The metadata (that is stored in the output) can later be used to exactly reproduce/understand your result, even if you have lost/forgot the command you used to create the file. This feature is present in all of Gnuastro’s programs, not just NoiseChisel.

The rest of the HDUs in NoiseChisel have data. So let’s open them in a DS9 window and then describe each:

$ astscript-fits-view xdf-f160w_detected.fits

A “cube” window opens along with DS9’s main window. The buttons and horizontal scroll bar in this small new window can be used to navigate between the extensions. In this mode, all DS9’s settings (for example, zoom or color-bar) will be identical between the extensions. Try zooming into one part and flipping through the extensions to see how the galaxies were detected along with the Sky and Sky standard deviation values for that region. Just have in mind that NoiseChisel’s job is only detection (separating signal from noise). We will do segmentation on this result later to find the individual galaxies/peaks over the detected pixels.

The second extension of NoiseChisel’s output (numbered 1, named INPUT-NO-SKY) is the Sky-subtracted input that you provided. The third (DETECTIONS) is NoiseChisel’s main output which is a binary image with only two possible values for all pixels: 0 for noise and 1 for signal. Since it only has two values, to avoid taking too much space on your computer, its numeric datatype an unsigned 8-bit integer (or uint8)32. The fourth and fifth (SKY and SKY_STD) extensions, have the Sky and its standard deviation values for the input on a tile grid and were calculated over the undetected regions (for more on the importance of the Sky value, see Sky value).

Each HDU/extension in a FITS file is an independent dataset (image or table) which you can delete from the FITS file, or copy/cut to another file. For example, with the command below, you can copy NoiseChisel’s DETECTIONS HDU/extension to another file:

$ astfits xdf-f160w_detected.fits --copy=DETECTIONS -odetections.fits

There are similar options to conveniently cut (--cut, copy, then remove from the input) or delete (--remove) HDUs from a FITS file also. See HDU information and manipulation for more.


Footnotes

(32)

To learn more about numeric data types see Numeric data types.