Segmentation is the process of (possibly) breaking up a detection into multiple segments (technically called objects and clumps in NoiseChisel). In deep surveys segmentation becomes particularly important because we will be detecting more diffuse flux so galaxy images are going to overlap more. It is thus very important to be able separate the pixels within a detection.
In NoiseChisel, segmentation is done by first finding the ‘true’ clumps over a detection and then expanding those clumps to a certain flux limit. True clumps are found in a process very similar to the true detections explained in Detection options, see Akhlaghi and Ichikawa  for more information. If the connections between the grown clumps are weaker than a given threshold, the grown clumps are considered to be separate objects.
The minimum area which a clump in the undetected regions should have in order to be considered in the clump Signal to noise ratio measurement. If this size is set to a small value, the Signal to noise ratio of false clumps will not be accurately found. It is recommended that this value be larger than the value to --detsnminarea. Because the clumps are found on the convolved (smoothed) image while the psudo-detections are found on the input image. You can use --checkclumpsn and --checksegmentation to see if your chosen value is reasonable or not.
Save the S/N values of the clumps into two files ending with _clumpsn_sky.XXX and _clumpsn_det.XXX. The .XXX is determined from the --tableformat option (see Input/Output options, for example .txt or .fits). You can use these to inspect the S/N values and their distribution (in combination with the --checksegmentation option to see where the clumps are). You can use Gnuastro’s Statistics to make a histogram of the distribution (ready for plotting in a text file, or a crude ASCII-art demonstration on the command-line).
With this option, NoiseChisel will abort as soon as the two tables are created. This allows you to inspect the steps leading to the final S/N quantile threshold, this behavior can be disabled with --continueaftercheck.
The quantile of the noise clump Signal to noise ratio distribution. This value is used to identify true clumps over the detected regions. You can get the full distribution of clumps S/Ns over the undetected areas with the --checkclumpsn option and see them with --checksegmentation.
Keep a clump whose maximum flux is 8-connected to a river pixel. By default such clumps over detections are considered to be noise and are removed irrespective of their brightness (see Flux Brightness and magnitude). Over large profiles, that sink into the noise very slowly, noise can cause part of the profile (which was flat without noise) to become a very large and with a very high Signal to noise ratio. In such cases, the pixel with the maximum flux in the clump will be immediately touching a river pixel.
Threshold (multiple of the sky standard deviation added with the sky) to stop growing true clumps. Once true clumps are found, they are set as the basis to segment the detected region. They are grown until the threshold specified by this option.
The minimum length of a river between two grown clumps for it to be considered in Signal to noise ratio estimations. Similar to --segsnminarea and --detsnminarea, if the length of the river is too short, the Signal to noise ratio can be noisy and unreliable. Any existing rivers shorter than this length will be considered as non-existent, independent of their Signal to noise ratio. Since the clumps are grown on the input image, this value should best be similar to the value of --detsnminarea. Recall that the clumps were defined on the convolved image so --segsnminarea was larger than --detsnminarea.
The maximum Signal to noise ratio of the rivers between two grown clumps in order to consider them as separate ‘objects’. If the Signal to noise ratio of the river between two grown clumps is larger than this value, they are defined to be part of one ‘object’. Note that the physical reality of these ‘objects’ can never be established with one image, or even multiple images from one broad-band filter. Any method we devise to define ‘object’s over a detected region is ultimately subjective.
Two very distant galaxies or satellites in one halo might lie in the same line of sight and be detected as clumps on one detection. On the other hand, the connection (through a spiral arm or tidal tail for example) between two parts of one galaxy might have such a low surface brightness that they are broken up into multiple detections or objects. In fact if you have noticed, exactly for this purpose, this is the only Signal to noise ratio that the user gives into NoiseChisel. The ‘true’ detections and clumps can be objectively identified from the noise characteristics of the image, so you don’t have to give any hand input Signal to noise ratio.
A file with the suffix _seg.fits will be created. This file keeps all the relevant steps in finding true clumps and segmenting the detections into multiple objects in various extensions. Having read the paper or the steps above. Examining this file can be an excellent guide in choosing the best set of parameters. Note that calling this function will significantly slow NoiseChisel. In verbose mode (without the --quiet option, see Operating mode options) the important steps (along with their extension names) will also be reported.
With this option, NoiseChisel will abort as soon as the two tables are created. This behavior can be disabled with --continueaftercheck.