Star Stretching to avoid bloated stars

[Peje]

I've been using pixinsight for a few years as my image processing software and I have always struggled with keeping star sizes small. I've tried numerous star reduction methods but I've never been happy with the dimmed fuzzy stars they give me. Some methods I can think of:

Stretching
MaskedStretch - Copying stars into HistogramTransformation strecthed image (Following Harry's tutorial)
HistogramTransformation - Using Lightness Mask

Post-Sretch Reduction
MorphologicalTransformation - Morphological Selection using starmask
MorphologicalTransformation - Erosion using starmask

Typically I have found that blending the MaskStretch stars in 50-50 with the HistTrans image gives me the best results, but I'm still not happy with the attached.

I have wondered if my equipment is the source of my problems, I have a 200mm f/4 Newtonian & an Atik 428 Mono CCD. This gives me a fast scope with a small field of view, pixel scale is 1.17 arcsecs. I have attached two images; one is of my auto-STF of my Lum stack, the other is the blended stretch I mentioned earlier.

My question would be, can anyone suggest other methods to try to allow me to produce nice tight stars without fuzzy halos?

Thanks in advance,
Pete

[Oddsocks]

Hi Pete.

It would be interesting to see a few uncalibrated and uncompressed lights in raw fits format from the same series to compare.

Although one would expect some loss of clarity in the image crops you have posted, overall the images seem very ‘soft’ to me. With your pixel scale I would have expected the diffraction spikes on that bright star in the lower left to be much crisper, especially from the Quatro with that very thin steel spider they use. The galaxy spiral arms also don’t seem that punchy.

Whether this is down to a processing/calibration problem, poor seeing or a problem with the raws from the equipment is impossible to say by looking at the stacked output, really need to see the raw data.

What do the stars look like in an integrated but otherwise uncalibrated series of lights when compared to the same series of lights but fully calibrated and which integration method do you use for the lights, Sigma clipping, Winsorized Sigma clipping, Averaged, Percentile clipping etc?

I too use PixInsight but have never needed to use star reduction in post, I have found a star mask or range mask applied before beginning any stretching is sufficient to control star bloat in most cases so I wonder if your stars are already bloated before you begin?

William.

[michael8554]

I agree, the stacked image looks out of focus, so a few subs to look at would be good.

Michael 

[Peje]

6 hours ago, Oddsocks said:

Hi Pete.

It would be interesting to see a few uncalibrated and uncompressed lights in raw fits format from the same series to compare.

Although one would expect some loss of clarity in the image crops you have posted, overall the images seem very ‘soft’ to me. With your pixel scale I would have expected the diffraction spikes on that bright star in the lower left to be much crisper, especially from the Quatro with that very thin steel spider they use. The galaxy spiral arms also don’t seem that punchy.

Whether this is down to a processing/calibration problem, poor seeing or a problem with the raws from the equipment is impossible to say by looking at the stacked output, really need to see the raw data.

What do the stars look like in an integrated but otherwise uncalibrated series of lights when compared to the same series of lights but fully calibrated and which integration method do you use for the lights, Sigma clipping, Winsorized Sigma clipping, Averaged, Percentile clipping etc?

I too use PixInsight but have never needed to use star reduction in post, I have found a star mask or range mask applied before beginning any stretching is sufficient to control star bloat in most cases so I wonder if your stars are already bloated before you begin?

William.

The link below is to subs from two nights, their equivalent calibrated images & the final stack of 237 subs.

https://1drv.ms/f/s!AlH5EVOP6MAJsyJ2t0ujPj2yDiwM

I've attached a screenshot of the integration settings I used for these, it is linear fit for sets with more than 25 subs. I got this from the 'Inside Pixinsight' book. I use batch preprocessing for the calibration and regristration then run ImageIntegration and finally DrizzleIntegration to give me the final stack.

I have noticed that my images never seem brilliantly sharp and diffraction spikes aren't overly dramatic, up til now I had assumed that this is down to me having pretty poor seeing most of the time over here in N.Ireland. If I'm doing something wrong then that would be great news.

A question about your stretching, are you saying you just appy a good starmask before you stretch the image? Is it an iterative process with different masks or just using one mask? I don't mind a long process to get good results so I'm open to anything.

[Oddsocks]

I had a look at all the images you provided Pete.

The calibration routine you are following seems ok. The main issue with this target is the rather soft focus.

Comparing your star from a single sub (left) with a similar star from one of my single subs (centre) shows how soft the diffraction spikes are in your image.

Though my image was taken at half the arc-sec/pixel resolution of yours and the telescope was an unflattened RC I would have expected a stronger diffraction spike in yours than in mine as the RC has relatively wide spider vanes which tends to suppress diffraction spikes.

The star on the far right is cropped from a completed Quattro image posted in 2017 by another SGL user @dph1nm and this is how I would expect the diffraction spikes to look in your final image.

You said that this has been noticeable in all your images since you acquired the telescope and suspect that this may be just poor seeing but even in NI you would have expected a few good nights a year. 

Assuming the OTA is fully acclimatised before starting the imaging run then something is not quite right.

You could try removing the flattener and taking a few images without it, the Quattro should be capable of much better than you are seeing at the moment and the flattener is where I would look first for possible issues.

All the stars in your linked images show some eccentricity, it could be tracking error or collimation, impossible to say from just looking at the image, though in my image there is tracking error both RA and DEC and still the diffraction spikes are well defined. Once you take the flattener out of the equation it will be easy to see if collimation error is the culprit.

I don't know how you focus but I found the Bahtinov mask to be too imprecise as focal ratio decreases and use the FWHM report from the image, the numbers are never subjective, just tweak the focus for min FWHM, adjust the exposure time as the chosen focusing star nears saturation with better focus, tweak focus a little more, repeat till done, though these days with automation I leave this to FocusMax.

Your question re masks:

I use multiple masks during image processing, starting with a star mask set for the brightest stars I give a small stretch to bring out the middle ranking stars, then create a new mask and adjust the parameters to cover the bright stars and the newly revealed stars, stretch a little more to bring out the faintest stars and then make a final mask to protect all the stars before working on the galaxy or nebulosity etc.

It can be quite time consuming to get the star mask parameters right, tweaking the structure growth and smoothness settings is necessary to cover the stars just enough so that processing artefacts are not introduced.

For working on the background and colour saturation I use either a range mask to cover all the stars and the galaxy or a star mask, depending what needs to be done.

In your Pinwheel example I would probably create and use as many as a dozen different masks during end-to-end processing.

I don't know how helpful any of the above has been Pete, I would try to resolve the basic image quality problem with the subs before trying to rectify the issues in post processing.

Hopefully another SGL member will also take a look at your linked images and see what they make of them.

William.

 

[michael8554]

Collimation ?

Michael 

[wimvb]

Reflectors will always have wider stars than refractors, due to diffraction. But to make them smaller:

1. Deconvolution in the linear stage. Use regularisation parameters to target only stars if you don't want to sharpen anything else.

2. Masked stretch.

3. Morphological transformation with a good contour mask. Use a small 3x3, sampling element. Morph selection rather than erosion. Follow up with mlt and a positive bias to brighten up the centre of stars. Use a starmask with low growth parameters.

[Peje]

5 hours ago, Oddsocks said:

I had a look at all the images you provided Pete.

The calibration routine you are following seems ok. The main issue with this target is the rather soft focus.

Comparing your star from a single sub (left) with a similar star from one of my single subs (centre) shows how soft the diffraction spikes are in your image.

Though my image was taken at half the arc-sec/pixel resolution of yours and the telescope was an unflattened RC I would have expected a stronger diffraction spike in yours than in mine as the RC has relatively wide spider vanes which tends to suppress diffraction spikes.

The star on the far right is cropped from a completed Quattro image posted in 2017 by another SGL user @dph1nm and this is how I would expect the diffraction spikes to look in your final image.

You said that this has been noticeable in all your images since you acquired the telescope and suspect that this may be just poor seeing but even in NI you would have expected a few good nights a year. 

Assuming the OTA is fully acclimatised before starting the imaging run then something is not quite right.

You could try removing the flattener and taking a few images without it, the Quattro should be capable of much better than you are seeing at the moment and the flattener is where I would look first for possible issues.

All the stars in your linked images show some eccentricity, it could be tracking error or collimation, impossible to say from just looking at the image, though in my image there is tracking error both RA and DEC and still the diffraction spikes are well defined. Once you take the flattener out of the equation it will be easy to see if collimation error is the culprit.

I don't know how you focus but I found the Bahtinov mask to be too imprecise as focal ratio decreases and use the FWHM report from the image, the numbers are never subjective, just tweak the focus for min FWHM, adjust the exposure time as the chosen focusing star nears saturation with better focus, tweak focus a little more, repeat till done, though these days with automation I leave this to FocusMax.

Your question re masks:

I use multiple masks during image processing, starting with a star mask set for the brightest stars I give a small stretch to bring out the middle ranking stars, then create a new mask and adjust the parameters to cover the bright stars and the newly revealed stars, stretch a little more to bring out the faintest stars and then make a final mask to protect all the stars before working on the galaxy or nebulosity etc.

It can be quite time consuming to get the star mask parameters right, tweaking the structure growth and smoothness settings is necessary to cover the stars just enough so that processing artefacts are not introduced.

For working on the background and colour saturation I use either a range mask to cover all the stars and the galaxy or a star mask, depending what needs to be done.

In your Pinwheel example I would probably create and use as many as a dozen different masks during end-to-end processing.

I don't know how helpful any of the above has been Pete, I would try to resolve the basic image quality problem with the subs before trying to rectify the issues in post processing.

Hopefully another SGL member will also take a look at your linked images and see what they make of them.

William.

 

Thanks William, I have removed the CC so hopefully will get some testing done tonight.

The rest of the info sounds great, I look forward to trying it

[Peje]

@Oddsocks

I've added a few more subs to that folder. I took three without the CC and am taking another 3 with the CC refitted. I don't notice a huge difference (other than the lack of coma) on regular stars and only a slight drop off of the diff spikes with the CC fitted. Does your eye see anything of interest?

Probably worth noting some haze blew in while I was refitting the CC.

 

I typically focus using a bahtinov mask first then let SG Pro run it's auto-focus routine while I watch, it used HFR instead of FWHM. There is the option to use FWHM but the user manual says HFR is generally the most accurate. I would be open to switching to FWHM but to be honest I think there's little to be gained.

Assuming my issue isn't the CC, I almost always collimate before starting a target so I would expect that to be good. In terms of collimation, I centered the secondary under the focuser at the start of the season and I use a hotech laser collimator to get the primary and secondary well lined up. I find that my Quattro holds collimation very well across temperature and general pointing position. One thing my system does suffer from is a bit of tilt in the imaging train, I'm hoping to cure that this summer with some tactically placed black tape to tighten up the fit of my 2" adaptor in the focuser tube.

 

On the masking, it all sounds reasonable. I spent quite a lot of time last year learning how to tune starmasks, mainly using the mask controls at the bottom, so making nice tight masks is fine. Going to give your method a whirl someday next week on my lunch.

[Oddsocks]

I have had a good look at the new additions Pete.

The star's airy disk shape in the non-CC images looks good and the disk is nicely round, the diffraction spikes are sharp and well defined. Although there aren't many stars to look at in the image they look just like I would expect from a fast newt on a camera with large pixels. Apart from the coma there is nothing wrong with these images at all.

In the 'with CC' images the star's airy disk is a diamond shape, or square, depending on your viewpoint, and the diffraction spikes are ill defined.

 

I am no optical expert but with the limited experience I have to date I would say the CC you are using is possibly faulty, not a good match for the telescope or a materials QA issue.  

The image the CC produces is very similar to what you would expect with a 'pinched' mirror. Maybe the lenses are too tight in the CC body.

Try loosening the CC's lenses retainer rings and then a gentle retighten, the lens retainers should just touch the lenses, if you were to shake or tap the CC the lenses should be able to move very very slightly, if they are held tighter than that then the lenses may distort leading to the problems I see with the star shapes.

If you try the above and it makes no difference and if the CC is still under warranty try getting it exchanged, or try a different CC. I seem to recall the Quattro's had a dedicated flattener for these fast newts so maybe finding a reasonable priced alternative is not going to be easy. Hopefully just a tweak to the CC lens retainers is all that is needed.

HTH

William.

[Peje]

3 minutes ago, Oddsocks said:

I have had a good look at the new additions Pete.

The star's airy disk shape in the non-CC images looks good and the disk is nicely round, the diffraction spikes are sharp and well defined. Although there aren't many stars to look at in the image they look just like I would expect from a fast newt on a camera with large pixels. Apart from the coma there is nothing wrong with these images at all.

In the 'with CC' images the star's airy disk is a diamond shape, or square, depending on your viewpoint, and the diffraction spikes are ill defined.

I am no optical expert but with the limited experience I have to date I would say the CC you are using is possibly faulty, not a good match for the telescope or a materials QA issue.  

The image the CC produces is very similar to what you would expect with a 'pinched' mirror. Maybe the lenses are too tight in the CC body.

Try loosening the CC's lenses retainer rings and then a gentle retighten, the lens retainers should just touch the lenses, if you were to shake or tap the CC the lenses should be able to move very very slightly, if they are held tighter than that then the lenses may distort leading to the problems I see with the star shapes.

If you try the above and it makes no difference and if the CC is still under warranty try getting it exchanged, or try a different CC. I seem to recall the Quattro's had a dedicated flattener for these fast newts so maybe finding a reasonable priced alternative is not going to be easy. Hopefully just a tweak to the CC lens retainers is all that is needed.

HTH

William.

Thanks William, it's well out of warranty. I also think I recall seeing it wasn't adjustable. I guess I will just have to save some pennies and orser the one SW designed to match the Quattro.

Do you think this is also causing my soft focus or would that be a separate issue? Or potentially you can't tell as the images dont have enough stars.

Thanks for your help.

Pete

[Oddsocks]

28 minutes ago, Peje said:

Do you think this is also causing my soft focus or would that be a separate issue?

I think this is most likely the reason for the soft focus, the lack of definition in the diffraction spikes suggests the photons are being "smeared" for want of a better term, and not being bought to a good focus.

Try a complete target acquisition without the CC, place a good bright galaxy in the image centre and when you post process just concentrate on that part of the image, ignore the coma stars, then compare to what you have been achieving in the past, if the CC is the problem you should see an immediate improvement in definition without the CC.

I use one of the Orion Optics (UK) AG8 fast newts and that uses a Wynne corrector, but that is not an option for the Quattro unless you changed the focuser too as the Wynne corrector is 76.5mm diameter (and costs £825 to buy separately from the scope!).

The images I have seen posted here on SGL taken with the Skywatcher Quattro dedicated flattener seem ok to me.

(I added a side-by-side image of the 'with' and 'without' CC stars to my last reply, just in case anyone else reading the thread wanted to see the differences without having to download the original images)

William.

[Peje]

On 26/04/2018 at 16:03, Oddsocks said:

Your question re masks:

I use multiple masks during image processing, starting with a star mask set for the brightest stars I give a small stretch to bring out the middle ranking stars, then create a new mask and adjust the parameters to cover the bright stars and the newly revealed stars, stretch a little more to bring out the faintest stars and then make a final mask to protect all the stars before working on the galaxy or nebulosity etc.

It can be quite time consuming to get the star mask parameters right, tweaking the structure growth and smoothness settings is necessary to cover the stars just enough so that processing artefacts are not introduced.

William.

 

I had a little fiddle on my lunch break today and while I was failing miserable at getting a mask to cover just the larger stars (probably due to doing two things at once), I was able to just use one of my 'normal' starmasks to just give the image a slight stretch (some small / middle stars coming up), then remove that mask and stretch as normal. Screen shot attached.

Whilst this isn't life changing, it is an improvement. I am using my worst data on purpose as it should exagerate any changes I make.

[Oddsocks]

3 hours ago, Peje said:

Whilst this isn't life changing, it is an improvement.

You have protected the core of the brightest star better in the latest image as well as tightening up on smaller stars, the old image shows an obvious ringing artefact in the core of the brightest star. I still think much of the 'fuzziness' is down to the flattener, it will be interesting to see how much difference it makes to a detailed target with no corrector, I believe the weather is forecast to be better in mid May and the Moon will be out of the way then too....

If you find getting a good star mask is difficult try a range mask instead, it is an easier tool to use because you can see the mask in real time using the live view window and tweak the settings to suit the image before committing to an 'apply'. It's a shame there isn't a manual mask editor in PixInsight, similar to how you would create a mask in Photoshop using a brush and an eraser tool, it would make PixInsight masks so much more user friendly.

 

[wimvb]

8 minutes ago, Oddsocks said:

It's a shame there isn't a manual mask editor in PixInsight, similar to how you would create a mask in Photoshop using a brush and an eraser tool, it would make PixInsight masks so much more user friendly.

Clonestamp? You can use it on multiple images. Eg, if you create a white image (PixelMath 1) and a black image (PixelMath 0), then you can use CloneStamp to paint black on the white image, or paint white on the black image.

[Oddsocks]

1 hour ago, wimvb said:

Clonestamp? You can use it on multiple images.

Thanks Wim.

I will have a look at Clonestamp, it's not something I have used before, I have plenty of old data here to experiment with.

William.

[Davey-T]

1 hour ago, Oddsocks said:

clThanks Wim.

I will have a look at Clonestamp, it's not something I have used before, I have plenty of old data here to experiment with.

William.

I  made a custom brush in PS to correct the dodgy stars in the corners of my Canon 24/105 lens.

Dave

[wimvb]

[Peje]

4 hours ago, Oddsocks said:

You have protected the core of the brightest star better in the latest image as well as tightening up on smaller stars, the old image shows an obvious ringing artefact in the core of the brightest star. I still think much of the 'fuzziness' is down to the flattener, it will be interesting to see how much difference it makes to a detailed target with no corrector, I believe the weather is forecast to be better in mid May and the Moon will be out of the way then too....

If you find getting a good star mask is difficult try a range mask instead, it is an easier tool to use because you can see the mask in real time using the live view window and tweak the settings to suit the image before committing to an 'apply'. It's a shame there isn't a manual mask editor in PixInsight, similar to how you would create a mask in Photoshop using a brush and an eraser tool, it would make PixInsight masks so much more user friendly.

 

I generally don't have much trouble with star masks, I was mutlitasking earlier so wasn't focusing properly. I should have just used binarise and then used convolution to soften the edges

[Tom OD]

Big improvement there in the New v Old image.

Good job.

Tom

[Peje]

On 30/04/2018 at 19:28, wimvb said:

I have fiddled nd fiddled but I cannot make it do what you did... could you give me a step by step?

EDIT: Not sure this is exactly what I'm trying to do. What I want is to copy a star (Dwarf Planet Ceres) from a sub, into a stacked image in exactly the same place as it was in the original sub. All images are aligned so perhaps I need to do something with Pixelmath and coordinates... another thing I don't know how to do!

[Davey-T]

In PS you could just put the stacked image on top of the sub as a layer alter the opacity to see both images and use the move tool to align them then change opacity back to 100% and the eraser to rub out the top layer to reveal the target.

Dave

[wimvb]

In PixInsight, you use the painting tool to create a mask. White where the item (ceres) to be copied is. Apply the mask to the image in which the new content should be placed, = target. Use pixelMath with the name of the image where you want to copy from, as the expression = source. Then apply this instance of Pixelmath to the target image.

If you don't know exactly where to put the mask hole, try this.

Clone the ceres sub. Dim it so no pixelvalues are 1.0, eg by pulling down the white point in curvestransformation to 0.8. Then create a white image as paint source. Apply a white clonestamp dot over ceres in the dimmed image. Use binarize at 0.9 to blacken the image, except where the white dot is. Blur with convolution. Now you have a mask which only targets ceres. Proceed as previous.

Btw, there's also a ReplaceWithPreview script in PixInsight, which may do something similar, but I've never had the need to try that.

[Davey-T]

6 hours ago, wimvb said:

In PixInsight, you use the painting tool to create a mask. White where the item (ceres) to be copied is. Apply the mask to the image in which the new content should be placed, = target. Use pixelMath with the name of the image where you want to copy from, as the expression = source. Then apply this instance of Pixelmath to the target image.

If you don't know exactly where to put the mask hole, try this.

Clone the ceres sub. Dim it so no pixelvalues are 1.0, eg by pulling down the white point in curvestransformation to 0.8. Then create a white image as paint source. Apply a white clonestamp dot over ceres in the dimmed image. Use binarize at 0.9 to blacken the image, except where the white dot is. Blur with convolution. Now you have a mask which only targets ceres. Proceed as previous.

Btw, there's also a ReplaceWithPreview script in PixInsight, which may do something similar, but I've never had the need to try that.

Blimey sounds complicated, it'd only take a minute in PS 

Dave

[wimvb]

2 hours ago, Davey-T said:

Blimey sounds complicated, it'd only take a minute in PS 

Dave

Explaining it takes longer than doing it