PixInsight recommendations

[MattJenko]

I have tried processing one of my images using PixInsight after downloading the trial this weekend. I didn't use it for calibration, I just followed Harry's tutorials for post calibration processing and have had a play with some random processes after that. It is a remarkable piece of software, but I would like a bit of a pointer as to what to do next, as until I get my Maths head on, I have no idea what a lot of the processes actually do in layman's terms, nor have a clue on how to use most of them.

This is 42 x 600 seconds with 10 dark, 10 flat, 10 bias and guided using an SW ED80/HEQ5 and a modded Canon 450D.

All suggestions welcome. I managed a lovely warm feeling image with this data using Photoshop (posted in a different thread and on Astrobin), but this version is starkly different! 

Thanks for looking.

[ollypenrice]

I only do a couple of things in PI before fleeing to the cosy warmth of Photoshop which is, to me, pipe and slippers to the cocaine and LSD of Pixinsight!!!

One thing you should do, though, is start by cropping before doing anything else. In the image above your uncropped edges have distorted the performance of DBE/ABE - or so I suspect. (Go on, tell me you didn't use DBE or ABE!!)

Olly

[MattJenko]

I did an initial crop, as this was a mixed set of exposures and there were really obvious stacking lines (done in DSS), but think i tried to cut it too fine as your eagle eyes have spotted. I did indeed use DBE, and liked it so much I tried it twice to see if I could equalize the remaining gradient. It is so much nicer and more precise than my duplicate image/blur/subtract image approach I use in PS!

I will try cropping some more.

I agree, Photoshop does seem like a high tea and scones compared to the mathematical bleakness of late night energy drinks and protein bars. I have to say I do like it though, although I am sure it cares not one jot about me in return

[NickK]

Hmm you've got dark rings around the stars from oversharpening sharpening.

[lensman57]

The cropping that Olly mentioned is crucial. The DBE tool has tried to flatten the background but because of the cropping there is a border visible which goes around the galaxy. You also have a lot of data in your stack of 42 subs and as a result the back ground should be as smooth as silk but in your Jpeg it is blotchy. Try and stretch the galaxy with the background and the  stars masked off. You should also run an instance of SCNR tool to remove the green cast after the final NR. I sometimes find it more effective if I import the image in PS and run the HLVG filter. PI is very complex and in particular the masking routines can be very time consuming so take your time, Rome wasn't built in a day.

A.G

[MattJenko]

Here is a more cropped version. It did make things easier, but I am still far too excited with deconvolution and sharpening, although I have discovered masks and de-ringing features...

Still loads to learn, but this is fun!

Thanks for advice

[Earl]

IMO one of of the big tricks with PI is to do little adjustments until you get used to whats its going to do. Undo then adjust the change, PI can overdrive many things very quickly.

The gradient on M31 is a pain to get right, and you need to build a careful map of points by hand, in my experiance it tends to over correct and give a harsh edge to the galaxy and not corecting the bacgrounds in the oposite cornes as you would expect.

On my last attempt on M31 I duplicated the Luminance (only got luminance done anyhow) messed with the levels to neutralise the background out then applied it as a mask to the original image to control the background without effecting the galaxy, it worked to some extent, (applying curves to the image with the mask inverted and normal helps see how your effecting the image, but make sure you cant see the mask to hide the red effect it has.)

Next problem with colour is getting the noise out of the background, shrink the starts a touch with a starmask and a morpholigial transformation (set it to do it in 12-20) might do it twice check Harry tutorials he covers it well.

I have attached my M31 so you can se what I did, I really need to get some colour and Ha for it, just not had the time or Moon in he right palce (or better still no moon LOL)

[ollypenrice]

Following Rogelio Bernal Andreo in Lessons From the Masters I've changed the way I use DBE. I now go for a  small number of markers and am very careful to keep them away from anything galactic or nebulous. Don't let them appear anywhere near genuine bright signal or it will give dark shadows round that object. It is best to study the stretched gradient map as well before applying it. It should be just that, a gradient map, largely featureless and, above all, it should in no way respond to the form of the main object. If you see any hint of, say, the shape of the galaxy in the gradient map you should scrap it and try again. You will also see edge artefacts at this stage if they exist. PI by default puts markers at the very edge. Moving them in a bit can sometimes help - or you can crop more ruthlessly.

Olly

[Earl]

Nice tip Olly

[harry page]

Hi

I always say its the quality of the samples , not the number of samples used ans olly advice is vey sound ( 99% of the time)

PI quiz time ( without looking now )  what is the min number of samples required to get DBE to work ?

Harry

[MattJenko]

And there was me about to post my latest revision which used a LOT of samples! Quiz failed.

I am going to post anyway, as a bit of patience following Earl's tip of simply trying out multiple iterations of each process to see how it works and how twiddling the knobs affects things (and also Harry's advice of 'experiment') was well worth it, especially for the lovely named AtrousWaveletTransform. I have tempered my earlier over-enthusiasm for sharpening and the results are better I think (to me). There are still issues with the cores and some magenta in the outer arms which I now think are a result of my DBE over-sampling!!

I will go back and start again with a DBE dominated evening.

[ollypenrice]

Hi

I always say its the quality of the samples , not the number of samples used ans olly advice is vey sound ( 99% of the time)

PI quiz time ( without looking now )  what is the min number of samples required to get DBE to work ?

Harry

Dunno, Harry, but from principle I can't see fewer than four being all that useful so that's my guess! One in each corner.

Olly

[harry page]

Hi

I am glad you are playing with pixinsight- but you pushed the last one to far

As for DBE ( as only olly wants to play) 3 is the answer and with a smooth gradient even 3 can work ( see attached)

Harry

[harry page]

Hi

BTW  I did not crop on purpose to show dbe will work fine as long as the samples are not in unaligned frames

Harry

[MattJenko]

I have scaled back the downtown, night-time Tokyo neon feel to it all, and gone a little more subdued. I also tried a lot less samples, but 3 wasn't quite enough ! I am still having problems with the cores, but all in all, today has been a very enlightening and entertaining day, and for that I am very thankful for for all the advice, it is really most appreciated, believe me.

[gnomus]

As a rank beginner, I hesitate to chip in, but I think overall this is the best of the ones you have posted.  Indeed it is quite spectacular.  The only thing that sticks out a little bit is the burnt out central core of Andromeda.  This didn't seem nearly so noticeable in your first image.  I wonder if combining just the central core of image 1 with the above (perhaps by using a mask in Photoshop) would result in the best possible image.  

[harry page]

Hi

The core is like this because of HDR wavlets , if you are not happy with this

apply a adjusted lum mask to feather the effects of hdr wavlets to give a more natural feel

Harry

[MattJenko]

I feel like I am getting a walk-through masterclass here (which I am), and I can only say thank-you.

Gnomus - thanks for the compliment. I could do it in PhotoShop, but the aim of the exercise for me here is not necessarily try and get the best image (although I am), it is to learn PixInsight. I am only a few days into my trial and I am already convinced it is worth an investment in both time and the £££ for the license. I just now need to practice practice practice, and I am feeling very lucky to have quality imagers and experienced processors like Harry, Olly and Earl comment on my progress and give me advice. It is what makes this forum quite special in my opinion.

It is remarkable what a bit of sleep can do, and I am feeling much more comfortable with the GUI and the mechanisms/ideas behind PI.

[andrewluck]

Some nice work especially for day 2 of the trial. I think at that stage I still had my head in my hands in despair. Worked out OK in the end though and I haven't lapsed into PS for a long time now .

Masked stretch is very useful for compressing the dynamic range and can reduce the need for HDRMultiscaleTransforms which can turn nasty very quickly.

Andrew

[lensman57]

I feel like I am getting a walk-through masterclass here (which I am), and I can only say thank-you.

Gnomus - thanks for the compliment. I could do it in PhotoShop, but the aim of the exercise for me here is not necessarily try and get the best image (although I am), it is to learn PixInsight. I am only a few days into my trial and I am already convinced it is worth an investment in both time and the £££ for the license. I just now need to practice practice practice, and I am feeling very lucky to have quality imagers and experienced processors like Harry, Olly and Earl comment on my progress and give me advice. It is what makes this forum quite special in my opinion.

It is remarkable what a bit of sleep can do, and I am feeling much more comfortable with the GUI and the mechanisms/ideas behind PI.

The core and the stars can be protected by creating a range mask in multiple stages ( to exclude the stars but keeping the galaxy body )  and using the fuzziness slider to protect the core and then applying a convolution filter to the range mask to soften the effect like you would apply a Gaussian blur in PS. This mask can then be applied to the galaxy both for HDR wavelets and either a global curves saturation or selective colour saturation. I am not a great fan of the masked stretch but it is useful for keeping the star sizes under control, this masked stretch can then be applied to the main image using an accurate star mask and using the Pixel math tool to transfer the stars to the main image, a bit advanced but not difficult.

A.G

[MattJenko]

Thanks AG, I hadn't quite twigged the possibilities of spending that level of effort on the mask, I had just been doing some histogram stretches to make it look 'ok'.

I actually am finding the whole PI process and methodology quite intuitive and a real pleasure to use, just the details of some (honesty check : most) of the processes is a serious dark art but with the previews its just a case of experimenting without understanding a lot of the maths, although I am quite keen to know what I am doing to my data! At this stage, its a case of working out what works before delving much deeper. My next step is to use PI instead of DSS.

[lensman57]

Thanks AG, I hadn't quite twigged the possibilities of spending that level of effort on the mask, I had just been doing some histogram stretches to make it look 'ok'.

I actually am finding the whole PI process and methodology quite intuitive and a real pleasure to use, just the details of some (honesty check : most) of the processes is a serious dark art but with the previews its just a case of experimenting without understanding a lot of the maths, although I am quite keen to know what I am doing to my data! At this stage, its a case of working out what works before delving much deeper. My next step is to use PI instead of DSS.

Merry Christmas Matt,

PI is very complex and you are correct in order to understand the function of the variables one needs to be a mathematician at least or a wizard.

A.G

[ollypenrice]

What the image now has, to my eye, is a very hot spot in the middle and then a zone around that which is not sufficiently bright, indeed hardly brighter than the mid-placed spiral arms. I find this kind of processing far easier to manage using layers in Ps. Sorry to be a heretic on this.

Olly

[MattJenko]

The hot spot and surrounding dark zone is certainly down to me playing about with the HDRMultiScaleTransform process and not really getting it right, or visually appealing enough. I completely hear what you are saying about Photoshop for this kind of activity, but I would hesitate to blame the tools in the hand of this particular workman . Once I get a moment to sit in front of a computer for a period of time rather than a quick internet fix in between family based drinking and sleeping I will have another go.

Merry Christmas to all by the way!

[ollypenrice]

What I do, promiscuous processor that I am, is do changes in PI which I then blend in as layers in Ps. No fundamentalism for me!

Olly