Dither?

[Nova2000]

Hi all

As my topic says, what's dither? What's the use of the feature? 

Rajesh 

[Geoff_L]

As a relative beginner, here's my understanding. An issue with long-exposure DSLR imaging is something I've heard called 'colour mottling' -- i.e. noise in the form of seemingly random pixels of colour. These to some extent depend on the CMOS array of the camera's sensor; so a particular pixel is be more likely to produce a spurious pixel of a particular colour and the mottle pattern is to some extent fixed. Dithering is a technique where the mount is moved a few pixels in each direction about a central point between frames. Thus the mottling pattern 'overlays' different parts of the image and so can be recognised as noise. When the resulting frames are stacked, much of the colour mottling noise can be removed (although I haven't yet managed to figure out whether this is 'self cancellation' or whether the stacking software does it).

HTH & clear skies

[ollypenrice]

It's very simple. In a single sub exposure the noise from your camera and the stars in the sky are in a fixed relationship with each other on the picture. If you move the camera slightly for the next sub the noise from the camera will be in a different position relative to the stars in the sky. You take many subs - at least 12 for a good result - and align them on the stars then combine them by averaging them. Because the stars (and nebulosity) have been aligned they reinforce each other. The noise, however, has not been aligned. It is randomly distributed and will be averaged out.

That is dither used for noise reduction. It can also be used for a very sophisticated technique called drizzle stacking. In some circustances, depending on your sampling rate, drizzle stacking can be used to find sub-pixel information and so produce an image of enhanced resolution. The process is complex and best explored through Google, I'd say.

Olly

[Ruud]

It's used a lot in printing. A fine pattern of mixed red and yellow dot will allow you to print orange, even if you have only red and yellow ink.

The German wiki page has good examples of how it is used: https://de.wikipedia.org/wiki/Dithering_(Bildbearbeitung)

It's used to create gradients and to solve problems with banding. Gif images are usually heavily dithered.

From Wiki. Top to bottom: original, 6-colour representation without dithering, three 6-colour variants of dithering (diffuse, pattern and noise)

[ollypenrice]

18 minutes ago, Geoff_L said:

As a relative beginner, here's my understanding. An issue with long-exposure DSLR imaging is something I've heard called 'colour mottling' -- i.e. noise in the form of seemingly random pixels of colour. These to some extent depend on the CMOS array of the camera's sensor; so a particular pixel is be more likely to produce a spurious pixel of a particular colour and the mottle pattern is to some extent fixed. Dithering is a technique where the mount is moved a few pixels in each direction about a central point between frames. Thus the mottling pattern 'overlays' different parts of the image and so can be recognised as noise. When the resulting frames are stacked, much of the colour mottling noise can be removed (although I haven't yet managed to figure out whether this is 'self cancellation' or whether the stacking software does it).

HTH & clear skies

As you say, dither can fight colour mottle from DSLRs. Colour mottle is a large scale noise pattern affecting many of these cameras. The scale of dither used to combat mottle is large. Tony Hallas recommends 12 pixels or more. For other kinds of NR a much smaller dither can be used and drizzle stacking can, if I remember, be applied from a sub-pixel dither.

Even using the simplest of average algortithms in the stack, take a rogue pixel which reads 30,000 ADU instead of 10,000. You dither so that it lands in a different place relative to the stars each time. If you guide perfectly and don't dither it will retain its rogue value. However, if averaged in a dithered stack of 12 it will have a final value of 11666. (10,000x11 plus 1x30,000 all divided by 12.)

That's an awful lot better than 30000! In fact the 'average' can be far more sophisticated, rejecting outliers using Sigma Clip etc.

Olly