Binning, Dithering and Drizzling


When using a dedicated astro camera it is possible to combine the individual pixel data into groups of 4, 9 or 16 pixels defined as 2x, 3x or 4x binning.

This will have two effects: it will increase the sensitivity but reduce the resolution.

2x binning  will halve the resolution but increase the signal by a factor of 4 so a decrease of  sqrt(4) = 2 in noise level.  A shorter total exposure will be needed for a given signal to noise ratio.  There will also be smaller files to store and process.  You may not need the full resolution that your sensor gives so I usually use 2x binning when imaging.

Whilst I am using  SharpCap to platesolve the captured images and home in onto the target object, I use 4x binning and also use a high gain as noise will not really be a problem.  This will minimise the exposure time to capture sufficient stars so that frames can be plate solved or enable one to quickly adjust the mount manually to centre an object in the frame.

When using a DSLR, reducing the captured image size does not appear to bin the pixels.  In a post processing program such as Photoshop reducing the image size will bin the pixels so reducing the noise in the image as seen in the set of images below.  The full size image (left) when reduced to half size (middle) does have a reduced noise level, but the half size image produced by the camera has the same noise level as the full frame image.


The sensitivity of pixels across the field of view is not uniform as the sensitivity of the green, red and blue pixels varies on scale sizes of ~20 – 40 pixels.  This means that if one imaged a totally blank sky without light pollution with the sensor staying perfectly aligned, the image would not be a uniform grey but suffer from what is called “colour mottling” – slight variations in colour across the frame.

A very bad example!

The way to avoid this to allow the image to move across the sensor during the imaging period and so average out the colour variations.  There are two ways to eliminate the colour mottling.

The first is not to auto guide, but to let the image slowly drift across the sensor so that the stars are not trailed.  This means that short exposures are necessary which is not such a problem with CMOS sensors in DSLRs or astro cameras. ( I typically use exposures from 30 to 60 seconds.)

Movement across the sensor will happen automatically if the mount is not perfectly polar aligned.  This is the approach that I take.  However it can give rise to “walking noise” which is very obvious in an image.

Zooming into the image one can see the effects of hot pixels streaking across the frame.

This was a particular bad example taken with a DSLR (so un-cooled) on a warm summer night.  There is one effect that helps reduce their visibility.  Stacking the frames will integrate up the stars, but not the hot pixels so their relative brightness will get less and one might simply be able to increase the black level and loose them in the background noise.  However, as seen below the most obvious hot pixel trails may still be seen.

These are not too difficult to remove in Adobe Photoshop or Affinity Photo. [Affinity Photo is a superb low cost program that can do virtually all that is required for use by astrophotographers.]

Use the “Select colour range” tool to select the hot pixel colour (red in this case).

Expand the selection by 2 pixels.

Paint the whole image (using a big brush) with the background colour (usually black).   

Repeat for any green or blue pixel trails.   One may then need to increase the black level to produce a passable image having removed the walking noise.

When Autoguiding

If autoguiding, allow the guiding program to make random movements in the telescope pointing between every few images.  Programs such as PHD guiding can carry this out.


This was originally developed for use with the Hubble Space Telescope and its first camera – “Wide Field & Planetary Camera 1” (WFPC1).

To fully sample an image (which is 2D) the Nyquist Sampling Theorum states that the sensor pixels must subtend an angular size that is one third the hoped for resolution.  The Hubble Space Telescope has a nominal resolution of 1/20th of an arc second = 0.05 arc seconds.  However, the pixels of the Wide Field & Planetary Camera 1 (WFPC1) had pixels which subtended 0.043 arc seconds.  So the images produced by the 2 m aperture telescope were well under-sampled.

By taking many images which are slightly shifted on the sensor, upsizing them by some factor (1.5, 2 or 3) and then aligning and stacking them, the resulting image will, perhaps surprisingly, overcome the sampling limitation and give a higher resolution image with, perhaps, an up to a 50% improvement.

Many smartphones can now use this technique as do some cameras which use sensor based image stabilisation such as the Panasonic G9: 

“By using sensor-shift stabilization technology the 20 Megapixel Panasonic G9 camera can use a High Resolution Mode to capture and compile eight separate images to create a single 80MP raw file. This system creates a more highly detailed and colour accurate image than a single shot alone can produce and can create an image with a 10,368 x 7,776 resolution.”

The noise level is also reduced as the result is the average of 8 images.  The scene must be stationary and wind can be a problem.

If our deep sky  images are dithered, Deep Sky Stacker can employ drizzling.

As can “Autostakkert!” when used to process lunar or planetary images.

I have used 1.5x Drizzle in Autostakkert! with some success in taking high resolution images of the Moon using “lucky imaging”. 

A near full sized version of this image having a resolution of ~0.7 arc seconds can be found here.