Technical details

Registering
Alignment
File Groups
Stacking
Stacking Methods
Drizzle
Comet Stacking
RAW images development process

Registering


Star detection
For each picture DeepSkyStacker will attempt to automatically detect the stars.  In simple terms, DeepSkyStacker considers that a star is a round object whose luminance decreases regularly is every direction, and whose radius is no more than 50 pixels.   Note that DeepSkyStacker will reject elongated star images which might occur if your mount isn't tracking correctly.   Once the star is detected its exact center is computed by fitting a Gaussian curve to the luminance.

DeepSkyStacker will only stack images that contain at least eight stars that are common between all light frames.   In practice this means that you should set the Star Detection Threshold in the Settings.../Register Settings or Register checked pictures/Register Settings/Advanced dialogue so that DeepSkyStacker detects 20 or more stars to stand a good chance of finding eight stars in common between all light frames.

Application of darks, flats and offsets before registering
If dark, flat and/or offset frames are checked they are automatically applied before the registering process.
With light frames containing a lot of hot pixels it is highly recommended to check dark frames in order to avoid false stars detections which may highly perturb the alignment.

Automatic detection of hot pixels
Optionally DeepSkyStacker tries to detect hot pixels during the registering process to avoid identifying false stars.
You should note that this option only works for monochrome images and RAW images in the Super-pixels, Bayer.Drizzle, bilinear and AHD interpolation modes.

Adjusting the star detection threshold
The star detection threshold is 10% by default (10% of the maximum luminance).

You can modify this threshold in the Advanced tab of the "Register checked pictures/Register Settings" dialog, or by selecting Settings.../Register Settings.

Reducing the Star Detection Threshold will result in DeepSkyStacker finding more (fainter) stars, on the other hand if you increase the threshold, then only brighter stars will be detected and so this will reduce the number found.

Setting the threshold so low that many hundreds of star are found will be counter-productive as there will be much more data to process for star registration and if too many stars are detected there is a greater chance of mis-registration.   You should probably aim for over 20-25 stars and no more than a couple of hundred or so.

If you have set the detection threshold low and DeepSkyStacker is still not finding enough stars because the image is underexposed, you can increase the image brightness by using the "Brightness" adjustment in the Raw/FITS DDP Settings.

If your images are noisy (as a result (e.g.) of Light Pollution) you may need to enable the "Reduce the noise by using a Median Filter" option.

To help you finding the best threshold for your light frames you can compute the number of stars that will be detected. To do this, DeepSkyStacker uses the first checked light frame and temporarily activates hot pixel detection.

You should note that this number is only a guide and that the real number of detected stars may vary if you have checked dark, offset and flat frames.

  

Registering results
The registering results (number of detected stars, position, luminance of each star) are saved in a text file which name is the name of the picture file with an .Info.txt extension.
Thus it is not necessary to register the picture again for a future stacking process.

Registering results and parameters
The registering results are highly dependant of the selected parameters (especially the raw development parameters).
If these parameters are modified it is necessary to register the pictures again.

Stacking after registering
 DeepSkyStacker can chain registering and stacking processes. You just have to give the percentage of the pictures that you wish to keep at the end of the registering process to start the stacking process. Only the best pictures will be used in the stacking process.
Thus it is possible to launch the full process then to go to bed knowing that after a good night sleep it will be possible to view the first results.

Alignment
Offsets and angle computing
 During the alignment process the best picture (the picture with the best score) will be used as the reference frame unless you choose another reference frame using the context menu.
All the offsets and rotation angles are computed relative to this reference frame.

The offsets are rotation angles are computed by identifying patterns of stars in the frames.
To put it simply the algorithm is looking for the largest triangles of which the side distances (and so the angles between the sides) are the closest.
When a sufficient number of such triangles is detected between the reference frame and the frame to be aligned the offsets and rotation are computed and validated using the least square method.
Depending of the number of stars a bisquared of bilinear transformation is used.

For more information about the algorithms which inspired the one used by DeepSkyStacker you can consult the following sites:
FOCAS Automatic Catalog Matching Algorithms
Pattern Matching with Differential Voting and Median Transformation Derivation

Automatic use of previously computed offsets and angles
DeepSkyStacker is saving all the transformations between a reference light frame and all the other light frames so that it is not necessary to compute them again if the registering info have not changed.
The info is saved in a file named after the reference frame (and in the same folder) with an .stackinfo.txt extension.

File Groups
File Groups may be used to simplify multiple nights on the same object file management by logically grouping files for each imaging session.
If you use only the Main Group DeepSkyStacker is working exactly like before the introduction of File Groups.

There are two kinds of File Groups: the Main Group and all the others groups.

You can create as many File Groups as you want knowing that a file may belong to only one File Group.
When you start DeepSkyStacker only the Main Group is available. As soon as you add a file to the last available group a new empty group tab is created.

Example:
You shot the same subject two nights in a row.
For each night you have a set of Light, Dark and Flat Frames but the temperature was not the same each night and the Dark Frames are not compatible and the orientation was slightly different so your Flat Frames are also different between the two nights.

To associate each Light Frame with the good Dark and Flat Frames you just need to put all the Light+Dark+Flat Frames from the first night in one File Group and all the Light+Dark+Flat Frames of the second night in another File Group.
Since Offset/Bias Frames are common to all the nights they must be put in the Main Group.

DeepSkyStacker will automatically associate the Light Frames of the first night with the Dark and Flat Frames of the first night and the Light Frames of the second night with the Dark and Flat Frames of the second night.
The Offset/Bias Frames from the Main Group will be associated with the Light Frames of the first and second night.

Stacking
Background Calibration
The Background Calibration consists in normalizing the background value of each picture before stacking it.
The background value is defined as the median value of all the pixels of the picture.

Two options are available.

It is important to check one of these options when using Kappa-Sigma Clipping or Kappa-Sigma Clipping Median methods to ensure that the pictures being stacked have all the same background value.

Automatic calibration of flat frames
The goal of the automatic calibration of flat frames is to equalize the luminosity differences between the flat frames before combining them into a master flat.

The first flat frame is used as a reference. The other flat frames are normalized to match the average luminosity and dynamic range of the first flat frame.

Automatic Detection and Removal of hot pixels
The goal of the automatic detection and removal of hot pixels is to replace hot pixels with a value computed from neighbor pixels.

First the very hot pixels are identified by an analysis of the dark frames (or the master dark frame if available). Every pixel which value is greater than [median] + 16 x [standard deviation] (sigma) is marked as a hot pixel.

For all those pixels the value in the calibrated image (after offset/bias subtraction, dark subtraction and flat division) is interpolated from the neighbor pixels.

Automatic Detection and Removal of bad columns
On some monochrome CCD chip some columns are either dead or completely saturated because of blooming created by hot pixels.

The detection and removal of bad columns may be used in these cases.
It automatically detects 1 pixel wide vertical lines that are either saturated or completely dead and deal with these lines as if they were hot pixels by interpolation their values from neighboring pixels.

Entropy-based Dark Frame Subtraction
Dark subtraction can be optionally optimized so that the entropy of the resulting picture (light frame minus dark frame) is minimized by applying a coefficient between 0 and 1 to the dark frame.
The main goal of this optimization is the possibility to use dark frames not taken in optimal conditions (especially concerning temperature).

For more information about this method you can consult the following document:
Entropy-Based Dark Frame Subtraction

Stacking process
The stacking process of DeepSkyStacker is very classical.
Step 1
Creation of the master offset from all the offset frames (with the selected method).
If more than one offset frame is checked, a master offset is created as MasterOffset_ISOxxx.tif (TIFF 8, 16 or 32 bit) in the folder of the first offset frame.
This file may be used as the only offset frame the next time.

Step 2
Creation of the master dark from all the dark frames (with the selected method). The master offset is subtracted from each dark frame.
If more than one dark frame is checked, a master dark is created as MasterDark_ISOxxx_yyys.tif (TIFF 8, 16 or 32 bit) in the folder of the first dark frame.
This file may be used as the only dark frame the next time.

Creation of the master dark flat from all the dark flat frames (with the selected method). The master offset is subtracted from each dark flat frame.
If more than one dark flat frame is checked, a master dark flat is created as MasterDarkFlat_ISOxxx_yyys.tif (TIFF 8, 16 or 32 bit) in the folder of the first dark flat frame.
This file may be used as the only dark flat frame the next time.

Step 3
Creation of the master flat from all the flat frames (with the selected method). The master offset and dark flat are subtracted from each flat frame. The master flat is automatically calibrated.
If more than one flat frame is checked, a master flat is created as MasterFlat_ISOxxx.tif (TIFF 8, 16 or 32 bit) in the folder of the first flat frame.
This file may be used as the only flat frame the next time.

Step 4
Computing of all offsets and rotations for all the light frames that will be stacked.

Step 5
Creation of the final picture by adding all the light frames with the selected method.
The master offset and the master dark are automatically subtracted from each light frame and the result is divided by the calibrated master flat, then if the option is enabled the hot pixels detected in the dark frame are removed and the value is interpolated from neighbors..

Step 6
When the Bayer drizzle is enabled, the three RGB components are normalized to avoid the information's holes.

Step 7
The resulting picture is automatically saved in an AutoSave.tif file which is created in the folder of the first light frame.

RGB Channels Alignment
When this option is enabled DeepSkyStacker attempts to align the three channels to reduce the color shift between the channels on the resulting image.
The main visible effect is that the stars are not anymore red on one side and blue on the other side.

Each channel is registered (the stars are detected) and a transformation is computed between the best channel and the two others.
The transformation is then applied to the two channels which is aligning them on the best channel.

Automatic use of previously created master files
The existing master files (dark, bias, flat and dark flat) created from a list of files are automatically used whenever possible as long as:
- The list of the files used to create them has not changed.
- The settings used to create them are not modified. This includes the combining method and parameters and the RAW or FITS DDP settings when RAW or FITS files are used.

A text file containing the parameters and the list of the files used to create the master image is saved in the folder of the master file.
The file is named after the master file name with a .Description.txt extension.

When the description is not matching the new settings the master files are automatically created again.

This feature is transparent for the user who only sees a faster processing because it is not necessary to create the master files again.

Using a Custom Rectangle
You can tell DeepSkyStacker to use a Custom Rectangle which will define the position and the size of the resulting image.

First, you need to preview an image by clicking on it in the list. You can select any image, but since you are defining the rectangle of what will be visible in the final image you should select the reference light frame (the one with the highest score or the one you decided to use as the reference light frame by using the context menu).

Then you just select on the image the rectangle that you wish to use as a Custom Rectangle.
When you will start the stacking process, the rectangle you just created will be selected by default as the stacking mode.

This option may be really helpful when used along with the Drizzle option that double or triple the size of the resulting image and thus is using much more memory and disk space during the stacking process.

Indeed, when a Custom Rectangle is used, DeepSkyStacker is only using the memory and disk space needed to create an image the size of the Custom Rectangle.

Stacking Methods
Average
 This is the simplest method. The mean of all the pixels in the stack is computed for each pixel.

Median
This is the default method used when creating the masters dark, flat and offset/bias. The median value of the pixels in the stack is computed for each pixel.

Maximum
This is and ultra simple method which should be use with a lot of care. The maximum value of all the pixels in the stack is computed for each pixel.
It may be useful to find what is wrong in a stack by exhibiting all the defects of all the calibrated images.

Kappa-Sigma Clipping
This method is used to reject deviant pixels iteratively.
Two parameters are used: the number of iterations and the standard deviation multiplier used (Kappa).
For each iteration, the mean and standard deviation (Sigma) of the pixels in the stack are computed.
Each pixel which value is farthest from the mean than more than Kappa * Sigma is rejected.
The mean of the remaining pixels in the stack is computed for each pixel.

Median Kappa-Sigma Clipping
This method is similar to the Kappa-Sigma Clipping method but instead of rejected the pixel values, they are replaced by the median value.

Auto Adaptive Weighted Average
This weighted average is adapted from the work of Dr. Peter B. Stetson (see The Techniques of Least Squares and Stellar Photometry with CCDs - Peter B. Stetson 1989).
This method computes a robust average obtained by iteratively weighting each pixel from the deviation from the mean comparatively to the standard deviation.

Entropy Weighted Average (High Dynamic Range)
This method is based on the work of German, Jenkin and Lesperance (see Entropy-Based image merging - 2005) and is used to stack the picture while keeping for each pixel the best dynamic.
It is particularly useful when stacking pictures taken with different exposure times and ISO speeds, and it creates an averaged picture with the best possible dynamic. To put it simply it avoids burning galaxies and nebula centers.
Note: this method is very CPU and memory intensive.

Drizzle
Drizzle is a method developed by the NASA for the Hubble Deep Field observations made by the Hubble Space Telescope.
The algorithm is also known as Variable Pixel Linear Reconstruction.

It has a wide range of usages among which it can be used to enhance de resolution of a stack of images compared to the resolution of a single image while preserving the characteristics of the image (color, brightness).

Basically each image is super sampled just before being stacked, like twice or thrice enlarged (it can be any value greater than 1 but DeepSkyStacker is only proposing 2 or 3 which are common values), then projected on a finer grid of pixels.

The result is that the size of the final image is doubled (or tripled) and that a small object that was occupying only a few dozens pixels, will be using twice or thrice the number of pixels and will be easier to post process.

What and when you need to use the drizzle option
Basically you need only to have a lot of images which must not be perfectly aligned (a few pixels drift is enough).
It makes no sense to drizzle if you have only a few light frames.

Drizzling is particularly good at enhancing the resolution so it is very useful when shooting small objects with a short focal length.

On the right is a example of an almost unprocessed M57 (shot using a 10", F/D 4.72 OTA with a Canon DSLR).

Usually M57 is very small, but by using a 2x drizzle and around 100 light frames the resulting image is twice bigger and the resolution is much better.

Mouse over to see the
   unprocessed M57 - No Drizzle (enlarged 4 times)
   unprocessed M57 - 2x Drizzle (enlarged 2 times)


M57 - No Drizzle

Side effects of drizzling
The main side effect is that the amount of memory and disk space necessary to create and process drizzled images is multiplied by the square of the Drizzle factor. Of course the time needed to create such images is also much longer.

For example, using a 2x Drizzle with 3000x2000 pixel images will create a 6000x4000 pixel image, which will need 4 times the memory and disk space size, and will be much longer to create.

When using the 3x Drizzle option, everything is multiplied by 9 (3 squared) and unless you have a very powerful machine and a lot of memory and disk space available you don't want to use this on classic DSLR images.

However, on some small images (like the one created by the first DSI and LPI cameras), it could make sense to use the 3x Drizzle option to enhance the resolution.

A good way to limit the increase of memory and disk space necessary to use the Drizzle option is to use a Custom Rectangle.

Drizzle and Bayer Drizzle
Although they are using two flavors of the Drizzle method, it is not recommended to use the Drizzle and Bayer Drizzle option together.
DeepSkyStacker is issuing a warning when you try to do so.

Comet Stacking
Comets are fast moving objects and when comet images are stacked together two things may happen:
- if the alignment between the images is made using the stars, the comet is fuzzy
- if the alignment between the images is made using the comet, the stars are showing trails.

Starting with version 3.0, DeepSkyStacker is adding two comet stacking options:
- Create an image aligned on the comet that will have star trails
- Create an image aligned on the comet and on the stars that will not have star trails.

Here is an example of the different stacking modes (mouse over the text to see the result)


Standard stacking
The comet position is ignored.
The comet is fuzzy and the stars have no trail.

Comet stacking : star trails
The comet position is used, the image is aligned on the comet.
Star positions are used to compensate any field rotation.

Comet and stars stacking : "star freeze effect"
The comet position is used.
A first stack is created to extract the comet from the background.
Then a second stack is created to freeze the star (the comet is subtracted from each calibrated and registered light frame before stacking it).
Last, the final image is obtained by inserting the comet back in the image.

 


If you plan to align the image on the stars, you don't need to do what is described in the following paragraphs since it's the default behavior.

What you need to do
Step 1: Register the comet center
DeepSkyStacker can not automatically detect the comet center in the light frames.

First, you must set the comet position in all your light frames. It is done only once.

To do this, just select a light frame in the list and using the Edit Comet Mode set the comet center.
If the comet center is too faint or too bright you can force DeepSkyStacker to accept any position by holding down the Shift key while positioning the comet center.

Then save the result by clicking on the Save Changes button in the toolbar.
If you don't DeepSkyStacker will ask you and you will have an option to save the changes automatically.

Once the comet position is set and saved you will see a +(C) added to the star count in the #Stars column of the list.

You must repeat the operation for each light frame.

Hint
 If the date/time of the images is accurate (like when using DSLRs and some CCD camera) you can sort the images by date/time and set the position of the comet only to the first and last light frames and reference frame (the one with the highest score if you have not used the context menu to force another frame).
DeepSkyStacker will then compute automatically (just before stacking) the position of the comet center in all the light frames in the time span from which the comet center is not set.
To do this it will use the elapsed time between the first image and each image to interpolate the position of the comet.

 Step 2: Select the stacking mode
This is done in the Comet tab available in the stacking parameters dialog.
The Comet tab is available only if at least two light frames (including the reference light frame) have a registered comet.
From this tab you can select one of the three available Comet Stacking options.
 
Mixing Comet and non Comet images
DeepSkyStacker can use images with a registered comet and images without a registered comet in the same stack.
This may be very useful to get a better signal to noise ratio on the resulting image especially in faint background details (a comet passing near a galaxy or a nebula for example).

Which stacking methods
If you are looking to create images with star trails, average is the best method.
In all others cases you should use Median stacking with small stacks and kappa-sigma with large stacks.

What results you can expect
Obviously the most demanding algorithm is the Comet and Stars stacking leading to the star freeze effect.
Slow moving comets are leading to hard to detect large objects or big stars and in this case the comet extraction process may be less than perfect.
In all cases, if you take a set of images from the same area without the comet (the day after or before) it will improve a lot the look of the final image.

RAW images development process
RAW files decoding
The RAW files created by the DSLRs are decoded using LibRaw Copyright © 2008-2019 LibRaw LLC, which is based on the original DCRaw by Dave Coffin.
DeepSkyStacker uses the latest version of LibRaw available at the time it was released, and will warn you if your camera isn't supported.

RAW files development process
A file is the equivalent of the digital negative. Thus, each raw file needs a development process.
There are two kinds of raw files: the one using a Bayer matrix (most of them) and the one not using a Bayer matrix (for example the one using a Foveon chip).

During the following explanations I will only consider the raw files created by a Bayer matrix based DSLR.

Bayer Matrix
 First, a small reminder of what is the Bayer matrix
When you are using an 8 mega-pixels DSLR, the CMOS or CCD chip is a black and white chip of 8 mega-pixels on which is glued a Bayer matrix which is in fact a pattern of RGBG or CYMK filters in front of each pixel (other patterns are possible)

In the case of RGBG filters a fourth of the pixels are capturing red, another fourth blue and the remaining half green.
So your 8 mega-pixels DSLR is producing pictures with 2 millions red pixels, the same amount of blue pixels and 4 millions green pixels.

Then how the DSLR is creating "true" colors pictures?
Very simply by interpolating the missing primary colors from neighboring pixels.

Color reconstruction using the Bayer matrix - Interpolation

The first way to reconstruct the colors from the Bayer matrix is to interpolate the missing primaries from the neighboring pixels.

A lot of different interpolation methods are available producing bad to good results (linear, gradient...) but all are degrading the quality of the final picture by guessing what the missing colors should be.

When each picture is slightly blurred by the interpolating process, the stacking of several pictures is loosing a lot of fine details.

If you still plan to use the interpolation method know that the software coming with your DSLR is probably the worst choice when it comes to preserving quality in astrophotography pictures.

If you were using it to transform your raw files to 16 bit TIFF files you will instantly see an improvement by using DeepSkyStacker.
 

Color reconstruction using the Bayer matrix - Super-pixel

With LibRaw it is possible to access the Bayer matrix before any interpolation. Thus it is possible to use other methods to reconstruct the true colors without guessing the missing primaries with interpolation.

The Super Pixel method does not interpolate but instead creates a single super pixel from each group of four pixels (RGBG).


Indeed each group of 4 pixels contains all the necessary information for each primary and the luminance.
This method is very simple and has the drawback (or is it a bonus) to divide the size of the resulting picture by 4.

The Super Pixel method is giving very good results when only a few pictures are available for the stacking process.

 

Color reconstruction using the Bayer matrix - Bayer drizzle

The  last method which was suggested by Dave Coffin uses the property of the stacking process to compute the true RGB values of each resulting pixel by using the "natural" drift existing between each picture.

When a large number of pictures is available and with its sub pixel alignment, DeepSkyStacker computes bit by bit the true values of each primary for each pixel, all without any interpolation.
 
After the stacking process another algorithm is used to normalize the RGB values to avoid information's holes.

The Bayer Drizzle method gives excellent results when a large number of light frames is available and when the guiding precision is greater than one pixel (which is almost always the case).