How much improvement does a 3 nm OIII filter provide over 6 nm?

Hi Tim,
I currently use a 6nm Lumicon OIII filter for my bi-color images. Although I'm new to bi-color imaging (I've done primarily H-alpha and LRGB all my life), I can tell you that my results under moderately light-polluted skies are very good - no gradients or lightened subs. I've also used it during 60% moonlit nights, too - same good results.

Hope this helps a bit.
Steve

In my Bortle 5 location, with an f/5.5 scope, I found a visible but subtle Ha and Oiii improvement in going from 7nm L-eXtreme to 5nm Antlia ALP-T - some of that may be down to the much reduced halos though.  I would expect further improvement going to 3nm but I don't feel the need to change again to L-Ultimate or similar right now. If I was in a significantly worse light pollution situation then I would certainly consider investing in the 3nm versions.

This article may be of interest  https://astrogeartoday.com/narrowband-filter-shootout-chroma-3nm-vs-astronomik-6nm/,  also this discussion:  https://www.cloudynights.com/topic/746279-comparison-of-3nm-nb-filters-vs-7nm-filters-%E2%80%9Cis-it-worth-it%E2%80%9D/

I image from Bortle 7 skies.  I used a Astrodon 5nm OIII filter.  Since the OIII is usually fainter, it always had poorer S/N than the 5nm Ha. When I move to a Astrodon 3nm, the background noise greatly decreased and the S/N increased.  At that time I was using a mid sized chip (Sony ICX694AL). When I moved to a larger chip ( Kdack KAF-16200 & Sony IMX571), I had to go to 2" filters. I continued to go for the 3nm OIII. I purchased a Chroma F3 3nm 2" filter. For the Ha and SII, I went with the Astronomik MaxFR 6nm. I went with the Astromnomiks because of cost and good reviews. They have worked out well. The Ha and SII 6nm blocks out light pollution well. I needed the F3 Chroma and MaxFR because I will be using them on a F 3 scope. The Baaders have terrible halos. 

I am going up to my dark sky location  next  week.  I have purcheced Astronomik 2" 12nm Ha and OIII for  use there. The skies there are Bortle 2 and there is no need for light pollution blocking, and I will be using the Ha &OIII as luminance for emission nebula. I went with the regular Astronomiks because I will be using a slower scope at F4.5.

Lynn K.

Many thanks for all the replies folks.    I can infer from the comments that - at the very least -  I will get quite a noticeable improvement that will also permit OIII data collection on more of the partly moonlit nights

Of course ks_observer must be mathematically correct with respect to SNR (x 1.41) improvement.  I would guess also a 2 fold boost in  the relative brightness of the OIII signal relative to stars - that being the other parameter that matters since the brighter the star the more troublesome the 'ghost' it leaves behind after applying starnet ?

My scopes are F4 and F5  so should get by with just the 'normal' 3 or 4nm filters (as opposed to the fast versions) .  Hmm eye-wateringly expensive - but probably worthwhile if I can sell the 6nm perhaps.

Tim

I fail to understand how reducing the total light in the 2 emission lines of the OIII (their distance is 6 nm so if the window is less than that you're going to loose some or all of one of the two) is going to be an advantage?

Arnie:
One needs to separate using narrow band filters under light polluted skies from imaging when the moon is up. O-iii filters don't block moonlight. Moonlight is broadband, so it radiates all across the spectrum. Moonlight will pass through an O-iii filter just fine, regardless of its bandwidth. What narrow band filters do is block artificial light pollution since that usually radiates at very specific bands in the spectrum, not associated with the H-a, O-iii or S-ii bands. In that case, narrower bandwidth will result in a darker background and better contrast. I can squeak by imaging H-a if the moon is quarter phase or less and 60 degrees or more away from the target. Any imaging with the 6nm O-iii when the moon is up produces gradients I'd rather not deal with.

My 2 cents worth,

Arnie

Right, but I think people worry too much that a moon will ruin images.  The last two narrowband images I made were during the three days around full moon. 

People dont need to be scared of the moon.  You can make a satisfactory image.  If I didnt image during the full moon that would cut my imaging by half or more...

Here are a few full moon images...

I use 5nm. I will gather 03 during a full moon.  Of course its always better with less light pollution (moon or manmade) but I don't let that stop me. 

If you are having problems with star size thats likely a processing problem. How you are handling the data. 

FWIW I have used 3nm and 5nm extensively with a variety of scopes.  The issues you are describing will be resolved better through post processing methods than getting a 3nm o3 filter. 

If you'd like to explore this you can go into detail on what you are doing and see if tgere is something to work on

Hi Chris.  Thanks.  I would be grateful for your comments and think that it is anyway a useful topic to air  Impressive pics under moonlit skies!

I use a different process according to the  target.  Here just concentrating on Type II nebulae.  Generally I go for the more naturalistic HOO palette - with 10% HA in the blue channel to allow for implicit H beta.  Accordingly I also prefer to remove the NB stars and completely replace them with RGB stars from a broadband OSC image.  However - having chosen to  do this - the most problematic part of the process* is then clean star removal - and especially spiky Newtonian ones.  The bit I really don't like is resorting to using clone stamp to clean up the starless image where large stars were incompletely excised -  arbitrary and probably introduces artifacts

 For Type II nebula where HA is dominant and also entirely overlaps the OIII signal  then I use the HA luminance  to define the final image.  I think that this is justified because as well as offering much better SNR the HA defines the detailed nebula shape all the way out to the ionization edge  - whereas the weaker OIII signal is usually less structured and occupies just core regions of the nebula around the hotter stars .  Using HA luminance also inherently reduces star size - even if it does somewaht downplay the brightness of blue stars versus red.   So in the processing method below the OIII  really just adds only its chrominance to localize it into the final image .  Probably an OK process for  e.g. the PACMAN and WIZARD nebulae  but not for something like the LION or CRESCENT?  Anyway the steps.. (using PIXINSIGHT)


1)  DBE background removal and MLT noise reduction applied to the linear OIII and HA images
2)  Morphological reduction of  the OIII stars (starmask and morph) 
3)  Linear Calibrate OIII image to HA image   (thereby boosing OIII -omiting this step probably gives a more 'honest' image in terms of  total HA v OIII emission)
4)  Apply an equal stretch to both (Histogram transformation)
5)*  Now star removal  and the creation of a starless image .. two variations
    a)   first use starnet to remove the stars and then pixmath to  compose a NB image (R 100%HA, G 100%OIII, B 90%OIII 10%HA)
    b)   first compose an NB image,  then LRGB  to transfer HA luminance to the NB image (reducing star size) and then starnet to remove stars

    (process b)  works best but in both cases blue/ green 'ghosts' and diff spikes are present which have to be cleaned away using clone stamp - after lin calib.            and stretch the OIII stars are again much bigger than the HA)
6)  Curves and noise reduction applied to starless image
7)  Starnet to extract the stars only from a mildly stretched RGB OSC image -  use curves to improve colour saturation etc.
8)  Pixmath e.g. max(RGBstars, NB image) to combine the RGB stars  with the NB starless image
9)  Optionally - further reduce binarize stars - apply curves, further noise reduction  or HA luminance again etc.

I usually get quite nice results eventually  with this approach https://www.astrobin.com/2cers7/B/   and  https://www.astrobin.com/m7tjub/B/
but they hide the fact that artifacts may be introduced around bright stars - especially in OIII - when you attempt to remove and  replace them.  It really is quite a clunky method. Hence I was interested  in getting a higher ratio of OIII signal to stars in the first place.  I suppose also that part of the problem is using a Newtonian - because of the diffraction spikes -  and also my desire to lose the NB stars -- i.e. no problem if you don't seek to replace them.

best wishes
Tim


PS .  Writing the above did make me reflect that the elephant in the room is really the fact that Starnet just doesn't work all that well - and especially with spiky stars.  So how to avoid it? An alternative solution to star replacement  might be to just add together an RGB image and an HA_luminance adjusted NB image  using the PI max function?   The nebula should be more intense in the NB image and the stars more intense in the RGB image - and thus generate the desired result - maybe?

Tim Hawkes:
Good point but I think that the 500.7 nm line is about 3X more intense than at 495.7. My 6 nm filter sits on 501 nm (varies according to F #) and so won't pick up the shorter wavelength band anyway

From what I can see it is closer to 1/2. All the more reason to use broader OIII filter, which I used and use to this day. BTW, Hbeta isn't equal to Ha scaled. It is different.

Hi Roger,

That would be right (with provisio on what stellar type is doing the excitation) for emission nebulae in the main (with few exceptions, IC434 for example) but for PNs this isn't necessarily true. As a picture is wortier than 1000 words you can have a look here, righthand side:

https://www.astrobin.com/full/rtb6q1/0/?nc=collection&nce=2688

I bought Chroma 3nm 50mm unmounted SII, Ha, OIII.  For backyard Astrophotography (Bortle 5.5) it was the best purchase I made.  I do not see any additional light pollution in my OIII subs.  As far as light pollution in my subs, SII, Ha and OIII all look pretty similar from my backyard (Bortle 5.5).  I also use them in Dark Sky (Bortle 3), they really go deep in dark sky.  
In fact, I would go so far as to say that for Backyard, I would give up total megapixels (camera) and do what I could to fit these filters in my budget, even if I had to image with 1 1/4" size filters.  They just make backyard so much more enjoyable.

Tim Hawkes:
1) DBE background removal and MLT noise reduction applied to the linear OIII and HA images
2) Morphological reduction of the OIII stars (starmask and morph)
3) Linear Calibrate OIII image to HA image (thereby boosing OIII -omiting this step probably gives a more 'honest' image in terms of total HA v OIII emission)
4) Apply an equal stretch to both (Histogram transformation)
5)* Now star removal and the creation of a starless image .. two variations
a) first use starnet to remove the stars and then pixmath to compose a NB image (R 100%HA, G 100%OIII, B 90%OIII 10%HA)
b) first compose an NB image, then LRGB to transfer HA luminance to the NB image (reducing star size) and then starnet to remove stars

(process b) works best but in both cases blue/ green 'ghosts' and diff spikes are present which have to be cleaned away using clone stamp - after lin calib. and stretch the OIII stars are again much bigger than the HA)
6) Curves and noise reduction applied to starless image
7) Starnet to extract the stars only from a mildly stretched RGB OSC image - use curves to improve colour saturation etc.
8) Pixmath e.g. max(RGBstars, NB image) to combine the RGB stars with the NB starless image
9) Optionally - further reduce binarize stars - apply curves, further noise reduction or HA luminance again etc.

A few things to consider:

Try to not work on stars on individual channels.  Running processes like MT changes the intensity of the stars and could make it more challenging to combine with stars from other channels. 

I also dont really like Linear Fit.  I used to use it to help make the OIII channel stronger, but now I'm creating an SHO and running PCC on it to balance the channels.  Then separate for custom PM blends. 

When you stretch your channels prior to star removal, etc... how much are you stretching?  Honestly, you really only need a very light initial stretch on the data for very effective star removal.   Also, I never run processes like HT and star removal, etc... on separate channels.  I always combine my channels in the linear stage and then stretch and manipulate stars, etc... while combined.  You are doing a lot of processing on individual channels and that may lead to the mismatches you are describing with halos, etc.. 

Food for thought...

Also Tim, I'd ditch MT for star reduction entirely.  With the available tools to remove stars and process the nebula and stars separately and then combine, I'm very happy with controlling star size through stretching rather than reducing a stretched image.  Less artifacts and more natural.  I'm using StarX and not Starnet, but I dont think there are any limitations with starnet for this type of processing. 

I looked at the images you linked, and I think the darker halos and any other artifacts you are worried about are entirely processing driven.  I think you can avoid getting a narrower filter to achieve the goals you are looking for.   Correct me if I am not seeing what you are talking about. 

Nice work by the way

Hi Chris.  Thanks for your helpful comments.  You were certainly correct that my main problem was in fact the processing.  Firstly I must apologise for putting  in rather cryptic  links to my two  images in without any clue as to what  was pertinent.  The image of the wizard was in fact done without use of starnet to replace stars  - and I thought looked basically OK .  The image  of the heart on the other hand was  using the starnet method and  seen in close there are artifacts around the brighter stars  where I had been struggling to use clone stamp to excise ghosts that starnet had left behind in the starless inage

The data set  that I am working on currently is another set of wizard data where I am trying and comparing different processing routes and in view of some of your comments.   Tentative findings thus far based on this current data set are ..

1) Yes agree.  Best to ditch use of MT altogether - at least until  perhaps cosmetically as a near final step
2)  Yes again - controlling star size just with the stretch looks better.  And PI provides many ways of doing it (masked stretch, local histogram equalisation and various combinations of LRGB and pixmath which can  be used to transfer luminance in e.g. star-selective ways)
3)  Starnet?  - in my particular case it really just doesn't seem work satisfactorily.   Even taking your advice to minimise the stretch before deploying the tool it still leaves  ghosts in the starless image -  which just become more obvious as you stretch back up again.  Diffraction spikes also leave behind traces of the 'wrong' colour also an uneradicated problem.  So for me Starnet is henceforth banned from my imaging - and especially using a Newtonian.  



For me a better way to put the RGB broadband stars into the image was via a combination of  the PI max function,  curves stretches and LRGB luminance transfer as detailed below.


4)  Agree about linear fit too  -  I have generated better images being more flexible about the OIII versus HA relative stretch .  Also I am now not doing anything (other than background removal and some MLT) on the individual channels


In the end have got to some images that look OK and a process (below) that looks to have worked quite well .... stiil not perfect with some haloing around the stars but better than where I was with needing to deploy the clone tool






In case anyone is still following this saga - and might want to do something similar  and uses PI - the starnet-free modified process was as below..

1) DBE background removal and MLT noise reduction applied to the linear OIII and HA images
2) Apply a lowish stretch to both (Histogram transformation) so that median signal of both is about equal and maximum nebula intensity is similar
3) DBE, PCC, MLT and SCNR also applied to a broadband RGB OSC image also stretched by histogram transformation
4) Combine the OIII and HA into a NB image using pixmath
5) Transfer luminance from the HA image to the NB image using LRGB. (this reduces and dims the stars back to the HA level and sharpens the nebula)
6) Stretch up the RGB image using curves so that the stars are clearly brighter than in the NB image and also sufficiently coloured (and the nebula is still relatively dim).  Use image statistics to check if necessary - and also look to minimise the background relative to the NB image
7) Now apply Pixmath function max(RGBimage, NBimage) to create a new image.  This new image should comprise the RGB stars plus the NB nebulosity.
8) Now reapply  the HA image luminance using LRGB. This step again reduces back down the size of the now correctly coloured RGB stars
9)  Use image statistics to measure the median signal and then apply a masked stretch (good stretch without blowing up the stars) and then (my preference depending on object) local histogram equalisation  to bring out faint contrast.
10) Finally curves , MLT for a final noise reduction step and (optionally) a masked star size reduction
 
Tim