Why does my M33 have a lack of blue?

[WolfieGlos]

A few weeks ago we had clear skies every night for 5 days, which was great as I recently upgraded from a star adventurer to a HEQ5 and this was the first time it was used…and bar a few polar scope calibration issues it worked flawlessly 😀

Using a Canon 800d (modded) + Evostar 72ED and no filters, I pulled in 85 x 240s subs on M33, dithered every 5 frames using APT and stacked with 50 bias, 60 flats and 8 darks in SiriL.

After stacking and processing in SiriL, Starnet and GIMP, attached is the best version that I could produce, and I’m fairly happy with it…aside from wanting more subs…!

However I have seen some great images on this forum recently of M33 where the galaxy has a blue cast to it, which now makes me doubt at what I’ve produced.

After many different processing attempts M33 always comes out grey with a yellow core. I’ve tried it with/without photometric colour calibration and background extraction but the same resulting colour comes out every time.

I also had a similar result with M31, and I didn’t question it, but now I’m thinking is it something I’m doing wrong?

[ONIKKINEN]

43 minutes ago, WolfieGlos said:

Using a Canon 800d (modded) + Evostar 72ED and no filters,

Does this mean no filters of any kind, even a UV/IR cut one? Also i guess the method of modding would affect how deep into IR the camera sees, if im not mistaken there are full spectrum mods and mods where the stock IR cut filter is replaced with an astronomy useful IR cut filter that passes to 700nm or so but not further.

The reason why a full spectrum pass is a problem is because most bayer matrix filters turn completely transparent somewhere around 800nm meaning that it turns monochromatic and will destructively affect your colours. Its essentially diluting your colours, specifically the blues since there could be a lot of near IR signal posing as blue in the capture.

Alternatively, something went wrong in processing. You could post the linear stack (before any processing has been done) and others could have a look if something went wrong.

[WolfieGlos]

23 minutes ago, ONIKKINEN said:

Does this mean no filters of any kind, even a UV/IR cut one? Also i guess the method of modding would affect how deep into IR the camera sees, if im not mistaken there are full spectrum mods and mods where the stock IR cut filter is replaced with an astronomy useful IR cut filter that passes to 700nm or so but not further.

The reason why a full spectrum pass is a problem is because most bayer matrix filters turn completely transparent somewhere around 800nm meaning that it turns monochromatic and will destructively affect your colours. Its essentially diluting your colours, specifically the blues since there could be a lot of near IR signal posing as blue in the capture.

Alternatively, something went wrong in processing. You could post the linear stack (before any processing has been done) and others could have a look if something went wrong.

Hi Onikkinen,

So the camera has been modified by Astronomiser to remove the rear UV/IR filter. Sorry I wasn't clear above, what I meant was that I was not using another type of clip-in or 2" filter, i.e. light pollution, dual/triband, etc. I live in a bortle 4 area, so skies are pretty good.

M31 and M33 are the only times I've used the 800Da for galaxies, and I decided to use it for M33 to try and capture some of the Ha. I've never imaged it before, and the last time I imaged M31 was a year ago when I was first getting into the hobby with a 77D and a 300mm lens, and even then I didn't get the blues...although that was processed in GIMP with a very limited knowledge of processing at the time. I do still have the stock 77D, which is useful as both cameras have the same size sensor.

I have attached the linear stack straight from Siril, with no processing if anyone would be kind enough to take a look 🙂 It will need cropping as you will see my framing wasn't perfect over the few nights I gathered the subs....

My workflow was as follows:

Photometric CC
Crop (x=97, y=0, w=5372, h=3585)
Canon Banding Reduction (amount=1.00)
Background extraction (Correction: Subtraction)
Deconv. (iter=100, sig=0.400)
Asinh Transformation: (stretch= 300.0, bp=0.00375)
Histogram Transf. (mid=0.261, lo=0.036, hi=1.000)
-- EXPORT TO STARNET V2, RELOAD STARLESS IMAGE
Generalised Hyperbolic Transformation (SF=1.313, LSI=1.592, SP=0.10219, BP=0.06425)
Deconv. (iter=1, sig=2.000)
Saturation enhancement (amount=0.20)
SCNR (type=0, amount=1.00, preserve=true)
Cosmetic Correction
Median Filter (filter=3x3 px)
CLAHE (size=16, clip=1.00)
Sharpening added to core in GIMP using masks, stars overlaid.

m33_result.fit

[michael.h.f.wilkinson]

I removed gradients and calibrated the star colours in Astro Pixel Processor, and got this result after a slight tweak in GIMP:

This is the colour I would expect without stretching saturation

[ollypenrice]

M33 is not terribly colourful if we are honest with the data. Hubble shows a warm central bulge consistent with old Population II stars but the image doesn't reach far into the spiral arms. Adam Block's Mount Lemmon image shows it a mild, silvery blue but his rendition is very colour cold by comparison with Hubble if we compare the bulges

Hubble  https://hubblesite.org/contents/media/images/2019/01/4305-Image.html

 

Mt Lemmon

 

Lots of imagers push for a decisive contrast between the bulge and this requires conscious processing efforts.

Olly

[michael8554]

I've done a fast-and-dirty tweak to the image in your first post:

I raised Luminance a little.

Blue levels were low, so I stretched blue.

Finally I raised Saturation.

Michael

[AstroMuni]

12 hours ago, WolfieGlos said:

After stacking and processing in SiriL, Starnet and GIMP, attached is the best version that I could produce, and I’m fairly happy with it…aside from wanting more subs…!

At the end of the day, if you are happy with the result then thats it 🙂  But yes, you could get more colour into it ... I played around with your processed image (jpeg) in Mac Preview and here is what I got

[ONIKKINEN]

11 hours ago, WolfieGlos said:

Hi Onikkinen,

So the camera has been modified by Astronomiser to remove the rear UV/IR filter. Sorry I wasn't clear above, what I meant was that I was not using another type of clip-in or 2" filter, i.e. light pollution, dual/triband, etc. I live in a bortle 4 area, so skies are pretty good.

M31 and M33 are the only times I've used the 800Da for galaxies, and I decided to use it for M33 to try and capture some of the Ha. I've never imaged it before, and the last time I imaged M31 was a year ago when I was first getting into the hobby with a 77D and a 300mm lens, and even then I didn't get the blues...although that was processed in GIMP with a very limited knowledge of processing at the time. I do still have the stock 77D, which is useful as both cameras have the same size sensor.

I have attached the linear stack straight from Siril, with no processing if anyone would be kind enough to take a look 🙂 It will need cropping as you will see my framing wasn't perfect over the few nights I gathered the subs....

My workflow was as follows:

Photometric CC
Crop (x=97, y=0, w=5372, h=3585)
Canon Banding Reduction (amount=1.00)
Background extraction (Correction: Subtraction)
Deconv. (iter=100, sig=0.400)
Asinh Transformation: (stretch= 300.0, bp=0.00375)
Histogram Transf. (mid=0.261, lo=0.036, hi=1.000)
-- EXPORT TO STARNET V2, RELOAD STARLESS IMAGE
Generalised Hyperbolic Transformation (SF=1.313, LSI=1.592, SP=0.10219, BP=0.06425)
Deconv. (iter=1, sig=2.000)
Saturation enhancement (amount=0.20)
SCNR (type=0, amount=1.00, preserve=true)
Cosmetic Correction
Median Filter (filter=3x3 px)
CLAHE (size=16, clip=1.00)
Sharpening added to core in GIMP using masks, stars overlaid.

m33_result.fit 277.14 MB · 6 downloads

Were you expecting something like this as a result?

Did very quick processing only, certainly not a full process. Looks like Siril photometry servers are down or something, i cant get it to platesolve and also cant get any of my other images to platesolve either (which previously of course have platesolved). At first i binned the image x2 in ASTAP to what i think is better sampling and higher SNR.

So the workflow for this image in Siril was:

1. Crop artifacts and generally for nicer framing
2. Background extraction
3. Manual color calibration with the core as white reference, works out ok for M33. Photometric CC would probably turn out a little bit different but cant do it now.
4. Asinh stretch at 1000, this greatly stretches colour while keeping the balance ok
5. Histogram transformation for the rest

In Photoshop:

• Selected background pixels only, desaturated and denoised them
• Selected signal pixels only, saturated them by +30
• NoiseXterminator
• Levels adjustment for blacks

Very simple processing only here, i think you may have tried too hard with your list of things, and have done them in incorrect order. Generally the siril workflow should always go in the 1-5 order shown here, with 4 and 5 optional, but i recommend doing the Asinh 1000 stretch first if you want colours to be nicer in the end. In many cases doing SCNR green is required, was not for this case. If it is required do it AFTER the Asinh 1000 stretch but before the final stretch so not every bit of green goes away. If you incoporate starless processing in this workflow, dont stretch the image fully in step 5, leave a bit of wiggle room for later.

You can drop deconvolution completely, the Siril tool only works with very high SNR images, which this (and most others) is not, and even then its not that good. Canon banding reduction should be done on the calibrated subs if necessary, its too late to do it on the stack as the banding is no longer horizontal or where it should be. I dont think its necessary for this image at all, you're not going to stretch it far enough for it to matter (at least i did not). The generalized hyperbolic doodad is great at killing colour, use at your own risk. Avoid if you dont really need it, like in this case where the dynamic range required to show the object of interest isn't really that difficult, an image with IFN or some faint nebula with background signal everywhere is another matter. The contrast limited histogram thing is not really something you need either, in this image or in general. Do fiddling with contrast in PS/Gimp later where you can place the adjustments exactly where you want to with layers. Also the median filter is a strange choice, you deconvoluted the image assumedly to emphasize detail, but then you went ahead and blurred it all out with the median filter? If the image is noisy, deal with it later or bin more.

There is a bit of chromatic aberration visible here in the brighter stars, i think you may want to consider placing an actual UV/IR cut filter somewhere in the imaging train. The astromod, while not full spectrum since only the rear filter was removed, probably passes a little bit too much deep purple and red.

Edited February 1 by ONIKKINEN

[alacant]

Hi

I think that Canon have reached a sweet spot with the 800d, although I'm surprised not to see more red bits. I suppose it depends upon what filter(s) were removed. Some nice detail emerging toward the centre of the galaxy and the 72 has done a good job controlling the stars.

Have a look at the colour balance maybe? Siril has an excellent colour calibration algorithm. Not a fan of processing involving star removal. Try without perhaps...

Cheers and HTH

Edited February 1 by alacant

[WolfieGlos]

Wow, thanks all !

 

1 hour ago, ONIKKINEN said:

Were you expecting something like this as a result?

Did very quick processing only, certainly not a full process. Looks like Siril photometry servers are down or something, i cant get it to platesolve and also cant get any of my other images to platesolve either (which previously of course have platesolved). At first i binned the image x2 in ASTAP to what i think is better sampling and higher SNR.

So the workflow for this image in Siril was:

1. Crop artifacts and generally for nicer framing
2. Background extraction
3. Manual color calibration with the core as white reference, works out ok for M33. Photometric CC would probably turn out a little bit different but cant do it now.
4. Asinh stretch at 1000, this greatly stretches colour while keeping the balance ok
5. Histogram transformation for the rest

In Photoshop:

• Selected background pixels only, desaturated and denoised them
• Selected signal pixels only, saturated them by +30
• NoiseXterminator
• Levels adjustment for blacks

Very simple processing only here, i think you may have tried too hard with your list of things, and have done them in incorrect order. Generally the siril workflow should always go in the 1-5 order shown here, with 4 and 5 optional, but i recommend doing the Asinh 1000 stretch first if you want colours to be nicer in the end. In many cases doing SCNR green is required, was not for this case. If it is required do it AFTER the Asinh 1000 stretch but before the final stretch so not every bit of green goes away. If you incoporate starless processing in this workflow, dont stretch the image fully in step 5, leave a bit of wiggle room for later.

You can drop deconvolution completely, the Siril tool only works with very high SNR images, which this (and most others) is not, and even then its not that good. Canon banding reduction should be done on the calibrated subs if necessary, its too late to do it on the stack as the banding is no longer horizontal or where it should be. I dont think its necessary for this image at all, you're not going to stretch it far enough for it to matter (at least i did not). The generalized hyperbolic doodad is great at killing colour, use at your own risk. Avoid if you dont really need it, like in this case where the dynamic range required to show the object of interest isn't really that difficult, an image with IFN or some faint nebula with background signal everywhere is another matter. The contrast limited histogram thing is not really something you need either, in this image or in general. Do fiddling with contrast in PS/Gimp later where you can place the adjustments exactly where you want to with layers. Also the median filter is a strange choice, you deconvoluted the image assumedly to emphasize detail, but then you went ahead and blurred it all out with the median filter? If the image is noisy, deal with it later or bin more.

There is a bit of chromatic aberration visible here in the brighter stars, i think you may want to consider placing an actual UV/IR cut filter somewhere in the imaging train. The astromod, while not full spectrum since only the rear filter was removed, probably passes a little bit too much deep purple and red.

Thanks for the detailed input and critique, much appreciated  

Yep, that's pretty much how I pictured it based on viewing other peoples work and considering the equipment I'm using. For a quick pass, that's really great! 

Sorry if this sounds silly, but what is binning the image x2? Is this a drizzle? I do have ASTAP but haven't used it for anything other than a quick mosaic trial and tilt checking (which I continually suffer from...but that's a different topic).

The Siril workflow is one I picked up somewhere a while back, it may have been here or on cloudynights before I joined any forums, and it's generally been OK but I do adapt it to suit what I have. That one was: Crop/Photo CC/BE/Deconv/Asinh/Histogram/Deconv/Saturation/SCNR. That one recommended Asinh +300, and then tweaking the black point until the point before the level slider reached 0 to avoid clipping. Did you adjust the black point at this stage?

I think I will give +1000 a go and see what it does. I thought SCNR was always required to remove the green cast, so again interesting that you didn't need to use it.

Deconvolution I generally don't use and I trial it each time I have an image, but this is one of the few times I ended up keeping the result. Using different values, the first run reduces star sizes and brightnesses (and also highlights, so I used GHT to try and bring them back...which now sounds silly...), and the second run sharpens the stars. Similarly, the banding reduction I rarely use but it seemed to make a difference here when I viewed it in Histogram mode in Siril. I can't believe I'm saying this again...but CLAHE I also rarely use!! Usually I adjust curves in GIMP and apply contrast this way.

The median filter, I'm not sure what it is actually supposed to do, but I find it reduces noise and without PS or PI I am currently very limited noise reducing options. I won't get a chance tonight or tomorrow, selfishly it's the wife's birthday, but I will take another shot at this on the weekend when I get a chance.

[WolfieGlos]

49 minutes ago, alacant said:

Hi

I think that Canon have reached a sweet spot with the 800d. Some lovely detail. Well done.

Just have a look at the colour balance maybe? Siril has an excellent colour calibration algorithm. Not a fan of star removal either. 

Cheers and HTH

Thanks very much  I was nearly going to get a 6D or 600D, but the reason I purchased the 800D was that it was a near match for my 77D which I already owned. They have the same battery (helpful!) and similar dynamic range/ISO/read noise results on photons to photos.

Stars...love em or hate em! 

[AstroMuni]

18 minutes ago, WolfieGlos said:

Sorry if this sounds silly, but what is binning the image x2?

In a nutshell its combining pixels to end up with a fatter pixel size. https://astronomy-imaging-camera.com/tutorials/everything-you-need-to-know-about-astrophotography-pixel-binning-the-fundamentals.html

22 minutes ago, WolfieGlos said:

I thought SCNR was always required to remove the green cast, so again interesting that you didn't need to use it.

As SCNR removes the green cast, in some cases you want the green to remain. So more of personal preference in most cases for me.

[ollypenrice]

4 hours ago, AstroMuni said:

At the end of the day, if you are happy with the result then thats it 🙂  But yes, you could get more colour into it ... I played around with your processed image (jpeg) in Mac Preview and here is what I got

Very magenta on my calibrated screen, suggesting green is too low. On the other renditions I'm seeing a slight excess of green. This is always the difficult axis in galaxies.

Since I post process in Photoshop, I measure the background using a 5x5 average sample and look for parity between the three colour channels. This is a helpful way to keep the lower brightness in order but the brighter parts are the difficult ones.

Olly

 

 

[AstroMuni]

3 minutes ago, ollypenrice said:

Very magenta on my calibrated screen, suggesting green is too low. On the other renditions I'm seeing a slight excess of green. This is always the difficult axis in galaxies.

And on mine too, but as I used the quick and dirty technique of trying to bring colour into the OP's image, that was to be expected 🙂

[ONIKKINEN]

1 hour ago, WolfieGlos said:

Wow, thanks all !

 

Thanks for the detailed input and critique, much appreciated  

Yep, that's pretty much how I pictured it based on viewing other peoples work and considering the equipment I'm using. For a quick pass, that's really great! 

Sorry if this sounds silly, but what is binning the image x2? Is this a drizzle? I do have ASTAP but haven't used it for anything other than a quick mosaic trial and tilt checking (which I continually suffer from...but that's a different topic).

The Siril workflow is one I picked up somewhere a while back, it may have been here or on cloudynights before I joined any forums, and it's generally been OK but I do adapt it to suit what I have. That one was: Crop/Photo CC/BE/Deconv/Asinh/Histogram/Deconv/Saturation/SCNR. That one recommended Asinh +300, and then tweaking the black point until the point before the level slider reached 0 to avoid clipping. Did you adjust the black point at this stage?

I think I will give +1000 a go and see what it does. I thought SCNR was always required to remove the green cast, so again interesting that you didn't need to use it.

Deconvolution I generally don't use and I trial it each time I have an image, but this is one of the few times I ended up keeping the result. Using different values, the first run reduces star sizes and brightnesses (and also highlights, so I used GHT to try and bring them back...which now sounds silly...), and the second run sharpens the stars. Similarly, the banding reduction I rarely use but it seemed to make a difference here when I viewed it in Histogram mode in Siril. I can't believe I'm saying this again...but CLAHE I also rarely use!! Usually I adjust curves in GIMP and apply contrast this way.

The median filter, I'm not sure what it is actually supposed to do, but I find it reduces noise and without PS or PI I am currently very limited noise reducing options. I won't get a chance tonight or tomorrow, selfishly it's the wife's birthday, but I will take another shot at this on the weekend when I get a chance.

There is a slight error in the recommended workflow you mentioned, and where i think your colour calibration took a wrong turn, and that is the PCC/BE order. Technically you could do it either way if both processes work perfectly, but in practice you should always feed PCC a gradient free image so that it can have the best chance to succeed. Here you first ran PCC and then removed the gradient, which also affects colours a little so you kind of undid the calibration.

With the Asinh 1000 i do adjust blackpoint a bit, but of course no clipping. The median filter option is a very crude way to denoise an image and will blur detail and noise in a 3x3 area, best to avoid and use denoising with Gimp somehow. Not sure how to as i dont use Gimp but there has to be a way, its such a common image editing feature that there is no way it cant be done (or a plugin that does it).

Binning is a bit of a minefield to discuss online, you get many different opinions. But binning is the opposite of drizzling in that you increase the size of a pixel opposed to decreasing it with drizzle. So if you drizzle x2 you half the signal to noise ratio and make the image 2x worse and 2x larger (with no extra detail in 99% cases), if you bin x2 you double the signal to noise ratio and lose no detail if the image was oversampled (or taken with a colour camera in most cases for complicated reasons, best leave it at that). Most people ignore all of this, up to you to decide whether its important to you.

[tomato]

Here is my take on boosting the blue on your excellent .fits file.

Basic workflow:

Crop and LP removal in APP

Image solve in PI then SPCC, BlurXterminator and StarXterminator. I then stretched the starless image and applied Curves Transformation, but I was not happy with the result. The SPCC/BXT file looked better when stretched in APP and saturation applied. This went back into PI for CT on the blue channel and SCNR on the green (@75%). NoiseXterminator applied and some subtle work in HST in Affinity Photo.

I personally like this level of blue, on reflection the M33 images I posted recently are too blue IMHO.

 

[WolfieGlos]

6 hours ago, ONIKKINEN said:

There is a slight error in the recommended workflow you mentioned, and where i think your colour calibration took a wrong turn, and that is the PCC/BE order. Technically you could do it either way if both processes work perfectly, but in practice you should always feed PCC a gradient free image so that it can have the best chance to succeed. Here you first ran PCC and then removed the gradient, which also affects colours a little so you kind of undid the calibration.

With the Asinh 1000 i do adjust blackpoint a bit, but of course no clipping. The median filter option is a very crude way to denoise an image and will blur detail and noise in a 3x3 area, best to avoid and use denoising with Gimp somehow. Not sure how to as i dont use Gimp but there has to be a way, its such a common image editing feature that there is no way it cant be done (or a plugin that does it).

Binning is a bit of a minefield to discuss online, you get many different opinions. But binning is the opposite of drizzling in that you increase the size of a pixel opposed to decreasing it with drizzle. So if you drizzle x2 you half the signal to noise ratio and make the image 2x worse and 2x larger (with no extra detail in 99% cases), if you bin x2 you double the signal to noise ratio and lose no detail if the image was oversampled (or taken with a colour camera in most cases for complicated reasons, best leave it at that). Most people ignore all of this, up to you to decide whether its important to you.

That's a very good point on the order of PCC/BE , noted - thanks. There is a noise reduction filter built into GIMP but it's quite limited and I find it usually destroys the stars and doesn't do a very good job, still leaving visible noise just less of it. I haven't paid for any software yet and that's probably holding me back in the final image quality. I did try a free de-noising program once, AstroDenoisePY, but it gave very poor results when I tried it.

[WolfieGlos]

3 hours ago, tomato said:

Here is my take on boosting the blue on your excellent .fits file.

Basic workflow:

Crop and LP removal in APP

Image solve in PI then SPCC, BlurXterminator and StarXterminator. I then stretched the starless image and applied Curves Transformation, but I was not happy with the result. The SPCC/BXT file looked better when stretched in APP and saturation applied. This went back into PI for CT on the blue channel and SCNR on the green (@75%). NoiseXterminator applied and some subtle work in HST in Affinity Photo.

I personally like this level of blue, on reflection the M33 images I posted recently are too blue IMHO.

 

Wow, that was from my fits file ?! 👏

Thanks for taking the time to do this as well Tomato, the detail you've pulled out of it combined with the lack of noise I'm really quite surprised by! I agree it's a nice level of blue, fairly subtle but obvious too. Further to my last post....definitely considering PI for the Blur and Noise Xterminator I've seen so much posted about lately.

One thing I've taken from this thread is that I thought I was getting the hang of processing....clearly I've still much to learn 😵

Edited February 2 by WolfieGlos

[tomato]

It’s all there in your fits file, these AI based tools aren’t for everyone, but they now form the mainstay of my processing workflow.

[wimvb]

Here's my attempt with PixInsight.

- Dynamic crop (removing stacking edges)

- DBE

- Background Neutralisation

-SpectroPhotometric Colour calibration

- Histogram transformation (levels in PS)

- Chrominance noise reduction (removing colour noise from the background)

- Curves Transformation, luminance and colour saturation

In my opinion, if you can get colour in the stars, you can get colour in the galaxy.

Edited February 2 by wimvb

[900SL]

There's a highly technical thread on cloudy nights regarding SiriL not accurately colour balancing with DSLR cameras from memory. Its way above my head but has involved the SiriL team and a couple of processing experts...

[ONIKKINEN]

6 minutes ago, 900SL said:

There's a highly technical thread on cloudy nights regarding SiriL not accurately colour balancing with DSLR cameras from memory. Its way above my head but has involved the SiriL team and a couple of processing experts...

Link? Would love to read it (with popcorn, naturally).

[wimvb]

6 minutes ago, ONIKKINEN said:

(with popcorn, naturally).

You asked for it: 😉

 

[900SL]

6 hours ago, ONIKKINEN said:

Link? Would love to read it (with popcorn, naturally).

Here you go, it's a lengthy read !

https://www.cloudynights.com/topic/841663-lost-color-in-siril-calibration/?fromsearch=1

[ollypenrice]

I think Wim and Tomato have it. A glory of warm core, blue arms and magenta splashes my look nice but I don't think it's the truth.

Olly