Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

AA filter, spatial filter and star colours

  • Please log in to reply
164 replies to this topic

#101 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 14 January 2019 - 01:24 AM

Adding a bit more to my earlier replies - the kind of imager (like me) who uses Bayer Drizzle to obtain greater resolution from an already pin-sharp image is the definitely the kind of user who will notice green or pink stars.  On the other hand an imager who takes Milky Way landscape pictures and only displays them at web size will (probably) never notice.  Most imagers fall somewhere between these two extremes.

 

There is an interesting pattern that has emerged every time Sony releases a new mirrorless camera.  Someone takes an astro-landscape image and makes a name for themselves by boldly declaring that this time Sony has completely fixed the issue.

 

e.g. the Sony A7R3: Sony a7R III: The Star Eater is No More (Petapixel)

 

followed one week later by:  The Sony a7R III Eats Stars: New Report (Petapixel)

 

The problem is usually that the original image shows a small amount of star trailing, imperceptible for a web-size image but sufficient to protect the stars from spatial filter damage.

 

Mark


Edited by sharkmelley, 14 January 2019 - 01:31 AM.

  • SandyHouTex and tkottary like this

#102 AlemanSky

AlemanSky

    Lift Off

  • -----
  • Posts: 5
  • Joined: 13 Jan 2019

Posted 14 January 2019 - 06:59 AM

Hallo Mark,

 

i've been following this thread with great interest and decided to create a program myself.

Plotting the value of a pixel against the maximum 24 neighbour yields the following chart for the RAW file provided by SandyHouTex:

DSC_0006_D750-k=5-max_.png

 

I don't understand the function you use to plot pixels against each other.

Can you help me out with the kernelfunction?

 

-Sergej


  • SandyHouTex likes this

#103 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 14 January 2019 - 02:02 PM

Hallo Mark,

 

i've been following this thread with great interest and decided to create a program myself.

Plotting the value of a pixel against the maximum 24 neighbour yields the following chart for the RAW file provided by SandyHouTex:

attachicon.gif DSC_0006_D750-k=5-max_.png

 

I don't understand the function you use to plot pixels against each other.

Can you help me out with the kernelfunction?

 

-Sergej

Hi Sergej,

 

I'm always very happy to see people trying to reproduce my results. 

 

The kernel function is very simple - a 5x5 box and I plot the centre value against the maximum of the 24 others.

 

I'm not sure what has happened in your plot but here are some suggestions.

  • I'm plotting the raw data values, which are 14bit integers.  They are therefore in the range 0-16383 and the bias level is around 600.  But your data is clipped at 0 and is in the range 0- 65000 or so.  This makes me think you are not plotting raw data values.  Your data has definitely been black subtracted. Maybe it has been debayered as well? I'm just guessing here.
  • I ignore the pixels near the border of the image where the 5x5 box touches the edge of the image.  The reason is that I don't want to guess what the Nikon algorithm does at the very edges of the image.

Meanwhile I will re-run my own data, just in case I have made an error.  There is always the possibility I have made a mistake - that's why I'm pleased to see folk checking my results!

 

Mark


Edited by sharkmelley, 14 January 2019 - 05:56 PM.


#104 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 24,608
  • Joined: 27 Oct 2014

Posted 14 January 2019 - 02:03 PM

The following is simply a report of my experience, not an attempt to disagree about anything.  OK?

 

I came across this thread and was baffled.  There are, what, 20 people with Nikon D5300/5500/5600s?  And nobody else reports this problem.

 

First I looked at an image of mine (cited below).  Chose something with a major starfield.  Could not see anything amiss.

 

So I looked at a light.  Debayered, stretched.  Some green stars.  Not all but a serious number, clearly not natural.  Almost certainly this problem, looked somewhat like sharkmelleys experiments.

 

Then I applied SCNR, default settings.  Surely a common thing to do with any one shot color camera.

 

The green stars vanished.  Not the stars themselves, just the green color.

 

Here's the final image.

 

https://www.astrobin.com/263253/H/

 

Now I could see why nobody else was reporting this.

 

I don't doubt that something is going on.  I do question whether it's a practical problem, and if it is for someone, for how many.  It's certainly not one for me, nor would I stop recommending this DSLR to beginners.

 

The real test would be for one imager to do two experiments, one with a D5300/5500, one as identical as possible, with a different camera that sharkmelley has found not to have the issue.  Do perfectly normal processing, nothing artificial.  See if there's any detectable difference in star color in the final image.  Not "green stars", just any detectable difference.


Edited by bobzeq25, 14 January 2019 - 02:39 PM.


#105 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 14 January 2019 - 05:54 PM

The following is simply a report of my experience, not an attempt to disagree about anything.  OK?

 

I came across this thread and was baffled.  There are, what, 20 people with Nikon D5300/5500/5600s?  And nobody else reports this problem.

 

First I looked at an image of mine (cited below).  Chose something with a major starfield.  Could not see anything amiss.

 

So I looked at a light.  Debayered, stretched.  Some green stars.  Not all but a serious number, clearly not natural.  Almost certainly this problem, looked somewhat like sharkmelleys experiments.

 

Then I applied SCNR, default settings.  Surely a common thing to do with any one shot color camera.

 

The green stars vanished.  Not the stars themselves, just the green color.

 

Here's the final image.

 

https://www.astrobin.com/263253/H/

 

Now I could see why nobody else was reporting this.

 

I don't doubt that something is going on.  I do question whether it's a practical problem, and if it is for someone, for how many.  It's certainly not one for me, nor would I stop recommending this DSLR to beginners.

 

The real test would be for one imager to do two experiments, one with a D5300/5500, one as identical as possible, with a different camera that sharkmelley has found not to have the issue.  Do perfectly normal processing, nothing artificial.  See if there's any detectable difference in star color in the final image.  Not "green stars", just any detectable difference.

It's interesting that you have seen the green star issue yourself.  There's no doubt that "green killer" processes such as SCNR will remove the green and make the stars look "normal" although not their true colour.   It's a happy coincidence that "green killer" is a frequently used process and it also tackles green stars.  It won't help for the D750's pink stars though!

 

No consumer camera is perfect for astrophotography.  You'll remember that when I did my review of the Nikon D5300 vs Canon 200D/SL2 that the Nikon outperformed the Canon in every category I could think of:

Canon 200D vs Nikon D5300 Conclusion

 

The spatial filtering was my only reservation and admittedly it's probably not going to be much of a problem to a beginner.  A beginner has plenty of other issues to get right and the D5300 makes a fine entry level camera.  But as the beginner grows in experience and becomes more of a perfectionist they might want to upgrade to an alternative camera.

 

Interestingly the subject of "green killer" processes also came up in that 200D vs D5300 thread.  My own view is that it is a quick and dirty fix for something that has gone wrong earlier in the processing sequence.  However, those who do use SCNR or HLVG as a standard part of their processing sequence (for whatever reason) won't notice the green stars either.

 

In the end, it all depends on one's level of experience and expectation.  For instance, green or pink stars will be a problem to those who want to use sophisticated processes such as PhotometricColourCalibration to set their white balance from a database of stars.  I've heard it argued that imagers with high expectation levels should buy dedicated astro-cameras instead.  I disagree.  If they choose the right consumer camera they will have a leading edge sensor at a much cheaper price.  In addition they'll be able to do astrophotography with a very portable setup - without laptops and power supplies.

 

Mark


Edited by sharkmelley, 14 January 2019 - 05:58 PM.


#106 fmeschia

fmeschia

    Surveyor 1

  • *****
  • Posts: 1,573
  • Joined: 20 May 2016
  • Loc: Mountain View, CA

Posted 14 January 2019 - 06:49 PM

I can testify that I was skeptical myself because I couldn’t see any trace of green stars in my actual astro images, but once I replicated Mark’s experiment under proper experimental setup (see this post), the issue became apparent.

 

So, the issue is there, and it’s software-based. Kudos to Mark for ferreting it out. We may debate its impact on real-world images, but not whether it exists. 

 

Francesco


Edited by fmeschia, 14 January 2019 - 06:51 PM.


#107 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 24,608
  • Joined: 27 Oct 2014

Posted 14 January 2019 - 06:59 PM

It's interesting that you have seen the green star issue yourself.  There's no doubt that "green killer" processes such as SCNR will remove the green and make the stars look "normal" although not their true colour.   It's a happy coincidence that "green killer" is a frequently used process and it also tackles green stars.  It won't help for the D750's pink stars though!

 

No consumer camera is perfect for astrophotography.  You'll remember that when I did my review of the Nikon D5300 vs Canon 200D/SL2 that the Nikon outperformed the Canon in every category I could think of:

Canon 200D vs Nikon D5300 Conclusion

 

The spatial filtering was my only reservation and admittedly it's probably not going to be much of a problem to a beginner.  A beginner has plenty of other issues to get right and the D5300 makes a fine entry level camera.  But as the beginner grows in experience and becomes more of a perfectionist they might want to upgrade to an alternative camera.

 

Interestingly the subject of "green killer" processes also came up in that 200D vs D5300 thread.  My own view is that it is a quick and dirty fix for something that has gone wrong earlier in the processing sequence.  However, those who do use SCNR or HLVG as a standard part of their processing sequence (for whatever reason) won't notice the green stars either.

 

In the end, it all depends on one's level of experience and expectation.  For instance, green or pink stars will be a problem to those who want to use sophisticated processes such as PhotometricColourCalibration to set their white balance from a database of stars.  I've heard it argued that imagers with high expectation levels should buy dedicated astro-cameras instead.  I disagree.  If they choose the right consumer camera they will have a leading edge sensor at a much cheaper price.  In addition they'll be able to do astrophotography with a very portable setup - without laptops and power supplies.

 

Mark

Pretty much agree, still have some questions.

 

One shot color cameras are very prone to excess "green", in the form of the ADU of the green channel.  If you look at post #1, you can see it in the Canon image.   That's why tools like HLVG exist.  I believe at least part of it may be to the fact that 50% of the pixels are green combined with the fact that the very broad "green" filter in the Bayer matrix routes a substantial amount of blue and red light to those green pixels.  Maybe most of the problem is.  Scroll down to the curve.

 

http://blog.teledyne...-color-cameras/

 

That's got to have an impact on "accurate" rendition of star color.  If our processing copes with that, maybe it will cope with this problem?

 

So my questions are.  What is the quantitative relationship between the DSLRs pervasive propensity to overemphasize the green channel, and this problem?  What qualitative factors might make it a problem beyond the quantitative aspect?

 

Opinion.  The following is not a matter of theory, it's just the way things are.

 

I think the intersection between the set of imagers who are sufficiently critical to have to worry about this particular issue, and the set of imagers who use DSLRs, is fairly small. 

 

There are a _great_ many reasons why almost everyone has some kind of computer at the scope now.   Even the great majority of DSLR imagers.   As the ranks of inexpensive cooled CMOS color cameras grows, and the availability of larger chips grows, the intersection above is likely to become smaller.

 

I have the RAW (NEF) data for that Andromeda.  It will be interesting to see if, indeed, Photometric Color Calibration is seriously disturbed.  I wasn't using it back then.  I'd take an even money bet PI copes.  At least to some degree.  <smile>  If it can deal with the inherent overemphasis of the green channel by DSLRs treating some red and blue light as green....


Edited by bobzeq25, 14 January 2019 - 07:15 PM.


#108 AlemanSky

AlemanSky

    Lift Off

  • -----
  • Posts: 5
  • Joined: 13 Jan 2019

Posted 14 January 2019 - 08:55 PM

Hallo Mark,

 

you're right with the clipped pixelvalues.

Unfortunately i used a somewhat bad library to read the raw files and haven't checked it properly.

Looking this evening into it i managed to reproduce your result :)

DSC_0006_D750.NEF-k=5-max_2000.png

 

Tomorrow i'll do these test with my cameras.


  • sharkmelley likes this

#109 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 15 January 2019 - 01:46 AM

Pretty much agree, still have some questions.

 

One shot color cameras are very prone to excess "green", in the form of the ADU of the green channel.  If you look at post #1, you can see it in the Canon image.   That's why tools like HLVG exist.  I believe at least part of it may be to the fact that 50% of the pixels are green combined with the fact that the very broad "green" filter in the Bayer matrix routes a substantial amount of blue and red light to those green pixels.  Maybe most of the problem is.  Scroll down to the curve.

 

http://blog.teledyne...-color-cameras/

 

That's got to have an impact on "accurate" rendition of star color.  If our processing copes with that, maybe it will cope with this problem?

 

So my questions are.  What is the quantitative relationship between the DSLRs pervasive propensity to overemphasize the green channel, and this problem?  What qualitative factors might make it a problem beyond the quantitative aspect?

 

Opinion.  The following is not a matter of theory, it's just the way things are.

 

I think the intersection between the set of imagers who are sufficiently critical to have to worry about this particular issue, and the set of imagers who use DSLRs, is fairly small. 

 

There are a _great_ many reasons why almost everyone has some kind of computer at the scope now.   Even the great majority of DSLR imagers.   As the ranks of inexpensive cooled CMOS color cameras grows, and the availability of larger chips grows, the intersection above is likely to become smaller.

 

I have the RAW (NEF) data for that Andromeda.  It will be interesting to see if, indeed, Photometric Color Calibration is seriously disturbed.  I wasn't using it back then.  I'd take an even money bet PI copes.  At least to some degree.  <smile>  If it can deal with the inherent overemphasis of the green channel by DSLRs treating some red and blue light as green....

DSLR cameras are not prone to excess green.  Otherwise how do they reproduce the colours on a colour chart?  The "star" colours in post #1 are because my LED light source is not white.  I could see visually that the LED torch produced a bluish/green colour.  I honestly don't know why some people have a problem with green in their processing and therefore need to use HLVG.  I think it is likely a problem with faulty background subtraction in their processing workflow.  On the other hand it might have something to do with this very issue we are discussing i.e. an artefact of the spatial filtering - I need to think about that some more.

 

Similarly the extended transmission to enable H-alpha makes little difference (unless the camera is used full spectrum).  Both the Nikon D810A and my modified Sony A7S do a pretty good job of reproducing the colours on a colour chart.  The white balance just needs to reduce the amount of red - that's how the D810A does it internally and it's how I manually do it.

 

You say that you think the intersection between the set of imagers who are sufficiently critical to have to worry about this particular issue, and the set of imagers who use DSLRs, is fairly small.  I think you are seriously underestimating people.  Give them the right information and then allow them to make their own choice.  Personally I was very discriminating right from my beginner days of imaging.  I had a choice between Canon and Nikon and I went with Canon because back then Nikon was infamous for their "star eater" (now copied by Sony!).  This predated the Nikon D7000 which was one of the first cameras to fix the star eater problem and it fixed it without producing green or pink stars either.  I always wanted to get good star colour: I was using arcsinh stretch almost from day one (using Christian Buil's IRIS processing software) so a camera producing green stars would have been a real turn off for me.

 

On a slightly different note, I'm looking very seriously at the Nikon Z6 as a possible imaging camera.  It looks like it might tick all the boxes.   Admittedly it's not an entry-level camera!

 

Mark


Edited by sharkmelley, 15 January 2019 - 02:00 AM.

  • tkottary likes this

#110 entilza

entilza

    Soyuz

  • *****
  • Posts: 3,799
  • Joined: 06 Oct 2014
  • Loc: Oakville, ON, Canada

Posted 15 January 2019 - 08:48 AM

Hi Mark, Thanks for your continued work on this topic, it's definitely interesting.

 

A friend of mine has a Nikon D5600 which performs quite well and haven't seen much on it, however I got some darks from it which I was wondering if you would be able to analyse the results?  If you need any more data I can try to get it.  Thanks again for your wonderful work.

 

https://drive.google...TARsb5f1ELlXnFX



#111 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 24,608
  • Joined: 27 Oct 2014

Posted 15 January 2019 - 09:58 AM

I honestly don't know why some people have a problem with green in their processing and therefore need to use HLVG.  I think it is likely a problem with faulty background subtraction in their processing workflow.  On the other hand it might have something to do with this very issue we are discussing i.e. an artefact of the spatial filtering - I need to think about that some more.

This is our one real disagreement.  Tony Hallas did not develop the Green Pixel Gun, nor Vicent Peris the extremely sophisticated SelectiveChromaticNoiseReduction, nor Ivo Jager provide the ability to map green to brown or yellow, because they (or others) are incompetent at background neutralization. 

 

The nature of the bandpass of the green filter in the Bayer matrix makes some green overemphasis inevitable.  The dumb "green" pixels cannot sort out the true green light they receive from the red or blue light they also receive.  Green speckles in the background are the most noticeable form of this distortion, but not the only one.

 

Minor point.  CCD/CMOS imagers are not immune from such things, the popular LRGB technique to reduce imaging time also distorts color, in a non-linear way.

 

We use a host of tools to manipulate color, and those manipulations always involve personal taste.  Some will find this particular color distortion a reason to avoid these cameras, based on reported experience, many (most?) will not.  They'll just manage it, as we always just manage color.  We do not simply seek out some cosmic reality.  <smile>

 

I'm surely not immune from these judgments.  My (and some others) aversion to broadband light pollution filters is largely due to the particular way those mangle color, ripping whole chunks from the spectrum.  The somewhat widely used CLS or tri-band filters are an extreme example.  Some people just cope with those, as well.


Edited by bobzeq25, 15 January 2019 - 10:02 AM.


#112 AlemanSky

AlemanSky

    Lift Off

  • -----
  • Posts: 5
  • Joined: 13 Jan 2019

Posted 15 January 2019 - 10:15 AM

Hi,

 

i run the test on my cameras:

DSC_1306.NEF-k=5-max.png

DSC_8122.NEF-k=5-max.png

The D700 and the D5000 are know for eating stars like children eat smarties.

 

Here are the charts of these cameras with the DarkCrurrentEnabler:

DSC_1307.NEF-k=5-max.png

DSC_8123.NEF-k=5-max.png

 

And here are the RAW files:

https://drive.google...tLzcvGI_ORxWdzq

 

I have also done a ISO bracketing test without the DarkCrurrentEnabler from Lo1 to Hi1.

The result was that the spatial filter is always applied with the same intensity regardless of the ISO setting.

 

For now i'm looking forward to do some real shoots of the milkyway for comparison (if the weathergod would let me).

 

-Sergej



#113 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 15 January 2019 - 04:34 PM

This is our one real disagreement.  Tony Hallas did not develop the Green Pixel Gun, nor Vicent Peris the extremely sophisticated SelectiveChromaticNoiseReduction, nor Ivo Jager provide the ability to map green to brown or yellow, because they (or others) are incompetent at background neutralization. 

 

The nature of the bandpass of the green filter in the Bayer matrix makes some green overemphasis inevitable.  The dumb "green" pixels cannot sort out the true green light they receive from the red or blue light they also receive.  Green speckles in the background are the most noticeable form of this distortion, but not the only one.

 

I think the "green" issue is sufficiently interesting that I want to try to understand what is going on, though I've rarely encountered it.  None of the potential explanations I have seen make any sense to me.  When you talk about the bandpass of the green filter, are you referring to the fact that the red, green and blue bandpasses overlap?  If so, it is designed that way - it's the only way a camera can record colour information that looks correct to the human eye. 

 

Anyway, yesterday I was actually thinking about this "green" issue and wondered whether or not the D5300 spatial filtering may be causing it.  So I did some calculations and couldn't come up with any mechanism.  Instead, what I did find was quite unexpected - I have a proof of why dark subtraction (in long exposure astro-imaging) is totally ineffective on the D5300 and on many other cameras.  I'll start another thread to discuss it because the purpose of this thread is star colour problems.  In any case I think most people are pretty bored with this thread by now smile.gif

 

Mark


Edited by sharkmelley, 15 January 2019 - 04:55 PM.

  • bobzeq25 and tkottary like this

#114 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 24,608
  • Joined: 27 Oct 2014

Posted 15 January 2019 - 06:12 PM

Yes, I'm referring to the fact that some red and blue light is translated to "green" signal.  The green filter bandwidth is wider than the red and blue, and it's right in the middle of the spectrum.  It's no wonder that special tools have been developed to control green.

 

The human eye is the joker in all this.  Especially the fact that human eyes differ substantially (and in a continuous fashion) in how they perceive color.

 

Looking forward to your explanation of why darks don't seem to work with my D5500.  The technical rigor you bring to these analyses is greatly appreciated by me, even if I don't always agree with your conclusions.



#115 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 15 January 2019 - 06:27 PM

 

Looking forward to your explanation of why darks don't seem to work with my D5500. 

Here it is:

Dark Subtraction and Spatial Filtering

 

Mark


Edited by sharkmelley, 15 January 2019 - 06:29 PM.


#116 Jerry Lodriguss

Jerry Lodriguss

    Vendor

  • ****-
  • Vendors
  • Posts: 7,796
  • Joined: 19 Jul 2008
  • Loc: Voorhees, NJ

Posted 16 January 2019 - 12:52 AM

DSLR cameras are not prone to excess green.  Otherwise how do they reproduce the colours on a colour chart?

 

I agree.

 

The thing I notice with folks who have problems with green is that they don't do color management correctly, starting with white balance.

 

And I always challenge these folks to process a normal daytime image with the exact same steps as their astro images, as a reality check to see if their colors are close to being correct.  And no one that I know if has ever done this with PI.

 

But they keep complaining about green.

 

These folks must not realize the amount of money that the daytime camera manufacturers put into color science.

 

Jerry


  • SandyHouTex likes this

#117 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 24,608
  • Joined: 27 Oct 2014

Posted 16 January 2019 - 02:01 AM

I agree.

 

The thing I notice with folks who have problems with green is that they don't do color management correctly, starting with white balance.

 

And I always challenge these folks to process a normal daytime image with the exact same steps as their astro images, as a reality check to see if their colors are close to being correct.  And no one that I know if has ever done this with PI.

 

But they keep complaining about green.

 

These folks must not realize the amount of money that the daytime camera manufacturers put into color science.

 

Jerry

The way you use white balance for color management simply compensates for the excess green, it's in no way proof that it doesn't exist.  Quite the contrary.

 

Also, you often use a CLS filter, which blocks a _major_amount of the spectrum that the Bayer matrix filter would otherwise let through.  In that case it's no wonder you wouldn't see excess green.  People can check it out.  The Bayer matrix curves are here

 

http://blog.teledyne...-color-cameras/

 

here's the CLS

 

https://rpubs.com/rr...omik-cls-filter

 

Even much of the red the CLS takes out is the same red that goes through the "green" filter of the Bayer matrix, causing the excess green signal.  Once again, a very effective way of reducing the excess green signal, if not a very good way of reproducing color. 

 

Bottom line, relevant to the thread.  Our equipment and our data capture techniques often distort color.  Color management is how we deal with that.  (The process always will be somewhat subjective.)  Based on the fact that some pretty good imagers with D5300/5500/5600 (I'm not including myself in the group) haven't found the subject issue to be a serious problem, it looks like it too is amenable to color management techniques.

 

If Jerry can make images with a CLS look good (and he can), people can do anything.  <grin>


Edited by bobzeq25, 16 January 2019 - 02:38 AM.


#118 Jerry Lodriguss

Jerry Lodriguss

    Vendor

  • ****-
  • Vendors
  • Posts: 7,796
  • Joined: 19 Jul 2008
  • Loc: Voorhees, NJ

Posted 16 January 2019 - 05:07 AM

The way you use white balance for color management simply compensates for the excess green, it's in no way proof that it doesn't exist.  Quite the contrary.

 

Also, you often use a CLS filter, which blocks a _major_amount of the spectrum that the Bayer matrix filter would otherwise let through...

 

Bottom line, relevant to the thread.  Our equipment and our data capture techniques often distort color.  Color management is how we deal with that.  (The process always will be somewhat subjective.)  Based on the fact that some pretty good imagers with D5300/5500/5600 (I'm not including myself in the group) haven't found the subject issue to be a serious problem, it looks like it too is amenable to color management techniques.

 

If Jerry can make images with a CLS look good (and he can), people can do anything.  <grin>

Not talking about using a CLS filter here though, the "folks" I was talking about don't use CLS filters, they have problems with greens unfiltered.

 

I think a better way to phrase it would be... I make whats left of the CLS-filtered spectral data look as good as possible with as correct a white point as possible and then maintaining that white balance throughout a color managed work flow all with the benefits of increased s/n from the filter.  Then I bite the head off a chicken in an ancient New Orleans VooDoo ritual, and I apply some special sauce. No magic there really. :-)

 

But this is beginner DSLR stuff. It is what everyone already is doing to their daytime pix too, right? 

 

Jerry


Edited by Jerry Lodriguss, 16 January 2019 - 06:27 AM.

  • SandyHouTex likes this

#119 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 16 January 2019 - 03:19 PM

Hi Mark, Thanks for your continued work on this topic, it's definitely interesting.

 

A friend of mine has a Nikon D5600 which performs quite well and haven't seen much on it, however I got some darks from it which I was wondering if you would be able to analyse the results?  If you need any more data I can try to get it.  Thanks again for your wonderful work.

 

 

Thanks for the data!  Here's the plot:

 

NikonD5600.png

 

All three channels look the same when plotted i.e. the red and blue pixels plotted against the max of their 8 same colour neighbours, the green pixels plotted against the max of their 12 green neighbours.

 

It's instructive to compare it against a plot of the D5500:

 

NikonD5500.png

 

You'll notice that the D5600 has no black clipping (i.e. no sharp horizontal cut off) because the data levels haven't hit that lower boundary and also there is some flexibility in the threshold used for capping the pixel values which allows some pixel values to remain slightly above the pink cut off line without being capped.

 

Overall though it's pretty much the same algorithm, which means small tightly focused stars will still tend to turn green.

 

Mark


Edited by sharkmelley, 17 January 2019 - 02:28 AM.

  • entilza likes this

#120 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 16 January 2019 - 03:48 PM

Hi,

 

i run the test on my cameras:

attachicon.gif DSC_1306.NEF-k=5-max.png

attachicon.gif DSC_8122.NEF-k=5-max.png

The D700 and the D5000 are know for eating stars like children eat smarties.

 

Here are the charts of these cameras with the DarkCrurrentEnabler:

attachicon.gif DSC_1307.NEF-k=5-max.png

attachicon.gif DSC_8123.NEF-k=5-max.png

 

And here are the RAW files:

https://drive.google...tLzcvGI_ORxWdzq

 

I have also done a ISO bracketing test without the DarkCrurrentEnabler from Lo1 to Hi1.

The result was that the spatial filter is always applied with the same intensity regardless of the ISO setting.

 

For now i'm looking forward to do some real shoots of the milkyway for comparison (if the weathergod would let me).

 

-Sergej

Hi Sergej,

 

That's a good set of plots!

 

I have a suggestion for the unhacked plots, which is to plot the values against the maximum of the 8 neighbours of the same colour, instead of 24 neighbours - like this:

 

NikonD700.png

 

When you do this you find that the chart for each of the 4 channels (R, G1, G2, B) is the same and it clarifies that for this camera there is no cross dependency between colour channels.  No pixel has a value greater than its 8 neighbours of the same colour.

 

This is the original (and infamous) Nikon star eater!

 

Mark


Edited by sharkmelley, 16 January 2019 - 03:49 PM.

  • AlemanSky likes this

#121 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 19 January 2019 - 06:58 AM

Thanks to Wei-Hao I now have some plots for the Nikon D800.  Here's the first plot, where I plotted each pixel against 16 neighbours - the 8 immediate neighbours plus the 8 other neighbours of the matching colour in the 5x5 block:

 

NikonD800.jpg

 

Notice that there are various obvious lines with different slopes.  This is a clear indication that the algorithm includes white balancing. If I white balance the data and re-plot using white balanced values, this is the result:

 

NikonD800_WhiteBalanced.jpg

 

Now the algorithm is clear - no (white balanced) pixel value is permitted to exceed the maximum of its 16 white balance neighbours.

 

Finally here's the version with hacked firmware:

 

NikonD800_HackedFirmware.jpg

 

This version has no hint of spatial filtering and there is no black clipping of the data.

 

Mark


Edited by sharkmelley, 19 January 2019 - 07:01 AM.

  • whwang, AnakChan, bobzeq25 and 1 other like this

#122 geethq

geethq

    Explorer 1

  • *****
  • Posts: 93
  • Joined: 09 Oct 2012

Posted 22 January 2019 - 05:52 AM

here are some modded(full spectrum)and unmodded darks of same D5300 if that helps. 5min 8min 10min single raw. also some master darks (20 stacked)

https://www.dropbox....xziYp_Xg2a?dl=0


Edited by geethq, 22 January 2019 - 06:20 AM.


#123 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 22 January 2019 - 02:01 PM

here are some modded(full spectrum)and unmodded darks of same D5300 if that helps. 5min 8min 10min single raw. also some master darks (20 stacked)

https://www.dropbox....xziYp_Xg2a?dl=0

Since there was no light reaching the sensor, how do you think the modded darks will be any different? wink.gif

 

Mark


  • AnakChan and fmeschia like this

#124 geethq

geethq

    Explorer 1

  • *****
  • Posts: 93
  • Joined: 09 Oct 2012

Posted 23 January 2019 - 02:40 AM

Since there was no light reaching the sensor, how do you think the modded darks will be any different? wink.gif

 

Mark

i personally took them separately due to change of power source. unmodded version don't have the battery inside. powered up with A/C adapter.

just posted to see if it has any help with your experiments as i mentioned above. i dont see any green stars in my final images anyway.



#125 sharkmelley

sharkmelley

    Aurora

  • *****
  • topic starter
  • Posts: 4,669
  • Joined: 19 Feb 2013
  • Loc: UK

Posted 23 January 2019 - 02:43 AM

i personally took them separately due to change of power source. unmodded version don't have the battery inside. powered up with A/C adapter.

just posted to see if it has any help with your experiments as i mentioned above. i dont see any green stars in my final images anyway.

Thanks.  I'll take a look just in case there are any differences.

 

Mark




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics