Star colours in near real-time viewing

[Martin Meredith]

I've become increasingly interested in open clusters since getting hold of RGB filters to use with my mono Lodestar. Having ascertained that the default settings for hue and saturation produce the best star colours (in the sense of matching the B-V colours reasonably well; see this thread), it is a real pleasure to turn the scope to some random cluster and watch the stellar configurations and colours appear. 

During the last session, largely inspired by posts from Jim Smith and suggestions from Dom543, I experimented with different sub lengths and brightness compression settings (linear versus arcsinh versus x^0.25). Shorter subs I felt would be less likely to result in saturated brighter stars (which produces white colours, with perhaps a tinge of the 'correct' colour around the edges where there is no saturation); but at the same time read noise might be bad enough to demand longer subs. Regarding brightness compression, I had no real idea what to expect.

I observed 3 clusters/suspected clusters: M29 and NGC 6996 in Cygnus, and NGC 6939 in Cepheus, close to the border with Cygnus. All three clusters were quite high (above 60 degrees) at the time, and there was no moon. Seeing was moderate, with StarlightLive reporting a FWHM of 2.8".

I'll add separate posts for each cluster as I manage to find star colour data to reference my captures. This post covers Messier 29 = NGC 6913.

In this first image I'm comparing the effect of sub length (5s, 10s or 20s). In each case I used 4 subs in each of RGB which were live stacked in StarlightLive. I'm also plotting as reference data V magnitude and B-V colour values from WEBDA, using the mapping from B-V colours to RGB defined on this site. B-V data was only available for a subset of stars, so there are more stars in my images than in the plots. Where more than one B-V estimate was available I computed the average. I'm using linear brightness scaling here.

At least up to 20s subs I don't see too much colour saturation -- just a greater magnitude depth.

The second image shows the differences between linear and the two nonlinear brightness scaling modes available in StarlightLive. In each case I adjusted brightness/contrast/black/white points to produce what I considered the most pleasing image -- not very scientific, I know. For me, pleasing means a fairly noise-free background and small stars. All were obtained with 20s subs i.e. the linear image is the same as the 20s sub case above.

There are clear differences between the 3 modes, with both x^0.25 and arcsinh producing softer edges to the brighter stars. There are also colour differences. I'm guessing that these result from the nonlinear compression affecting the R, G and B channels differently (i.e., the relative values of R, G and B might be altered following nonlinear transformation). However, I'm not totally sure about this given that some kind of channel-based histogram alignment is also carried out by StarlightLive.

To my eyes, both nonlinear modes reproduce the blues in the two leftmost bright stars better than the linear mode. As I get hold of data for the other two clusters I'll post equivalent images to see if this is also the case there.

What do others thing produces the best colour rendition?

BTW I'm aware that there are different points of view on what 'real' colour means and that is something else we could discuss...

Martin

 

[Jim Smith]

I think my vote is for X^0.25. On my monitor the white stars look white in that one.

[Martin Meredith]

Here's NGC 6939, which can be found about half a degree from NGC 6946 (the Fireworks Galaxy) and occupies a similar area. Here I didn't vary the exposure -- these are 4 x 5s subs in each of RGB. First, the full plot

and a zoom of a subregion just below left of centre where there is a nice arc of stars:

In both cases I'm plotting down to mag 17.0 since this seems to match the depth of the images.

Colours are definitely more muted in the nonlinear plots. I can't see much difference between x^0.25 and arcsinh in this case. However, the nonlinear plots to my eye appear to emphasise the blue channel.

Here's the full colour chart for this cluster.

NGC6939_det.pdf

Martin

 

[Ain Soph Aur]

Even with premium Astrodon Gen2 LRGB filters, I have had star bloat with the red filter when stacking with equal exposure subs with SLL and carefully focusing with each filter using TheSkyX @focus2 routine. I have tried using less red subs but not pleased with the results. This past weekend I experimented with equal numbers of RGB stacks but using 10% shorter red subs and and think I am on to something. Planning on experimenting with 5% shorter red subs as time permits, at 10% less red (27s) on 30s Green and Blue, I am seeing just a tad too much blue. 

[Dom543]

Hi Martin,

Very interesting study !

Since we  are all on new territory here, I am spelling out aloud here what and how I am thinking. Stars are the brightest objects in astronomy, so we are dealing here at the bright end of the tone scale. (I entirely ignore the faint end of the scale for the purpose of this post.) Non-linear options re-map the brightest tones to fainter ones. They do this to reduce blowout and to enable the user to increase overall brightness without excessive blowout. In other words, applying a non-linear tone map is only one step in the process towards getting a more pleasing image. It is not the only or the final step. It dims down certain elements of the image to make it possible to increase brightness, through increased exposure time, without getting too large areas to reach maximum saturation.

In light of this, your captures displayed with the non-linear options have accomplished the first step. The brightness of the brightest elements of the images, which are the stars, have been reduced. But as I see it, the purpose of this is to enable longer exposures. And hence more bit depth and color depth.

In line with this thinking, I believe that the image that would be a fair comparison to the linear option would be not one that is merely dimmed down by the use of one of the non-linear tone maps. But one, that has then be brightened back up to the same level as the liner map, by an extended exposure.

This is how I see it. I would be curious to hear your and others' opinion on this. Unfortunately, I will be travelling in the next couple of days and will probably not be able to log in to the forums.

Clear Skies!  --Dom

[RobertI]

Another very interesting study Martin.  

For M29, for the linear mode, the blue stars look most accurate at 5 seconds but the yellow stars look most accurate at 20 seconds. For the non-linear I think X^0.25 is the most accurate colour rendition, but both non linear are biased towards blue, with arcsinh being the most blue. 

Overall I think the 10 seconds linear conveys the most accurate colour when taking into account blue and yellow. 

For NGC6939 I found it more difficult, I think the yellows are better in the linear but blues are not so different....starting to get colour blindness!

As an aside, although not the most accurate colour rendition, I think the non-linear modes produce the most visually pleasing star shapes, despite the soft edges the stars seems smaller and exaggerated star spikes are nice, but not to everyone's taste I realise. 

[RobertI]

Curiously there is a fairly bright star on the image(s) which is not on the map. On the screen shot below I have ringed in blue the stars which match on the map and image and ringed in red the rogue star. A variable star perhaps?

[RobertI]

Looking at some other images of the same cluster, I think the rogue star is just missing from the map.......

[Martin Meredith]

Thanks for your comments so far. I won't reply in detail yet as I am still searching around for information to explain what is going on. Meanwhile, I've come across a couple of very useful resources:

1. Roger Clark's collection of web pages on astrophotography

It is worth reading all of section 2 which contains some interesting material on stellar colour and some explanations as to why certain images appear blue; plus some important advice on channel histogram equalisation (don't!) and automatic white balance (don't!), as well as an image showing the effect of 'a S-curve applied to boost contrast and saturation' on colours (don't use it!).

This agrees with my inclination not to engage in any (live) processing that alters the balance of the RGB channels without good reason if the idea is to maintain natural colours (we can argue about what that means!).

2. A detailed textbook Multicolor Stellar Photometry kindly made available via Astronomy Data Services at Harvard. This looks like an excellent resource for anyone wishing to understand the basis of the BV and other photometric systems, and covers factors such as atmospheric and interstellar extinction.

An interesting exercise is to do a web image search for a given cluster, and compare the results; you'll typically see a surprising range of colour variation even for the brighter stars (e.g. yellow-white stars appearing blue). What I'm hoping to achieve here is some simple live processing (or ideally absence of processing) that provides consistency within and across sessions, and a modicum of realism (red giants/carbon stars appearing red, young blue stars appearing blue/blue-white, in a range of pastel shades). It would be great to find an appropriate sun-like star in each image to perform a calibration but I recognise that this is a vain hope!

Martin

 

 

[RobertI]

2 hours ago, Martin Meredith said:

 

Thanks for your comments so far. I won't reply in detail yet as I am still searching around for information to explain what is going on. Meanwhile, I've come across a couple of very useful resources:

1. Roger Clark's collection of web pages on astrophotography

It is worth reading all of section 2 which contains some interesting material on stellar colour and some explanations as to why certain images appear blue; plus some important advice on channel histogram equalisation (don't!) and automatic white balance (don't!), as well as an image showing the effect of 'a S-curve applied to boost contrast and saturation' on colours (don't use it!).

 

What an amazing resource, I've read as far as section 2b, and learned a huge amount about astro-colours already. The graph has beaten me though, I can't understand what it is trying to convey. 

[Martin Meredith]

It is a great resource and I don't know why it has taken so long to find it. I've started at the beginning (section 1) as it is full of interesting (and surprising) information/opinions. Which figure in section 2b are you referring to?

Martin

[RobertI]

12 hours ago, Martin Meredith said:

It is a great resource and I don't know why it has taken so long to find it. I've started at the beginning (section 1) as it is full of interesting (and surprising) information/opinions. Which figure in section 2b are you referring to?

Martin

This is the graph...

y

[Martin Meredith]

Hi Rob

My reading is that the graph is showing that the colour we see is based on several factors: the 'true' distribution of stellar energy across wavelength, the absorption in the atmosphere, and the sensitivity per wavelength of the human eye. Multiplying the three components together provides a composite spectrum which is then summed across each of the wavelength regions corresponding to R,G and B, resulting in a final colour.

Martin

[Martin Meredith]

Here's another capture from last night, again comparing the effects of the different forms of compression. It shows a pair of open clusters in Lacerta near the border with Cepheus (this region is full of OCs). These are almost touching, but it is thought that King 9 is at more than twice the distance (26000 vs 11000 LY) and much older (3162 vs 447 MYr). There's a photometric study which includes both clusters here. There is a clear colour contrast between the two clusters in the shot.

This is with linear brightness scaling and 6 x 10s subs in each of RGB.

Here's a comparison of (top to bottom) linear, x^0.25 and arcsinh compression for the region containing the clusters. 

The star shapes are 'tighter' in the two nonlinear modes, but the colour balance has definitely shifted. Enough to reduce the clarity of the contrast between the two clusters? I think so. It may not be obvious in these shots but based on some captures I'll post later I see a consistent pattern, where arcsinh produces bluer colours than x^0.25.

Martin 

[Martin Meredith]

Globular clusters are also good test cases for the effect of linear versus nonlinear compression on colour rendition. Don Rudny and others have shown that nonlinear compression has a  substantial positive impact on globulars; I would say it is virtually essential for improving resolution. So what is the effect on colour?

Here are shots of M92 (top) and M13 with the best possible adjustments I could make live using linear (left), x^0.25 (middle) and arcsinh (right) compression. In each case I used 4 x 5s subs in each of RGB. This is quite a large image (5M).

here's a zoom to the cluster regions (1.6M)

I think this example shows the dilemma: it is impossible to get tight stars using the linear mode, and what's more, the stellar saturation in that mode is likely to be affecting the colours; on the other hand, in the nonlinear modes (particularly arcsinh) we get a colour shift, but the star sizes are far to be preferred. For my tastes the x^0.25 compression has done a good job but I'd like to get some kind of external validation of the colours here, particularly the yellow/blue distinction. 

I should add that in all these tests I've unchecked the automatic channel alignment option (although it doesn't make a great deal of difference to the histograms for these examples).

Martin

 

 

 

 

 

[stepping beyond]

A very good read for someone like me who is getting his feet wet in long exposure rgb , It's been a slippery muddy slope.