Large Carina Mosaic now with 'Round' Stars

[avandonk]

This is the result of four contiguous areas taken with the Canon 5DH and 300mm F2.8L. Gradients are a major problem with widefields but with my new methods managed to almost eliminate them without compromising the real image data too much.

Full resolution 5MB version
http://avandonkbl.bigblog.com.au/data/2/13839/image/mos52233420081207074918.jpg


Full resolution 13MB version
http://avandonkbl.bigblog.com.au/data/2/13839/image/mos132233520081207074917.jpg

The image covers 12.8 x 8.2 degrees. Zoom in and have a look around.

Bert

[robgreaves]

Great work Bert! Fun to pan around the large image and explore.

I have the prime 300mm L series too - I should give it a try sometime!

Regards,
Rob

[dugnsuz]

Very nice Bert.
"...my new methods", very mysterious!!
Doug

[avandonk]

I have mentioned it before Doug. This is a fuller explanation. GradientXterminator is fine if you have a galaxy or nebula in the center of your frame. You are in deep trouble with faint nebulae all over a widefield. Sure you get rid of gradients but you also lose faint nebula detail as it looks like gradients to the program.

What I do is split the colour image into RGB fits and then make R,G & B tifs of each. PS has not got a clue about fits images. I then treat each monocolour tiff with GradientXterminator in the usual way.

The advantage now is you can isolate real faint nebulosity from gradients and GX does not fart around with the other colours. I do not know what his algorithms or protocols are but this eliminates unwanted suppression of real faint nebulous data. Each image has its own method that you must work out by trial and error. It pays to be consistent with the RGB for the same image for Detail and Aggressiveness and only vary the areas with the Lasso tool.

Hope this helps. Any futher questions?

Bert

[dugnsuz]

Thanks Bert,
I have GradXterminator so I may give this a go.
Do you need a plug in to split the RGB image into fits files?
I'm using PS CS2.
Doug:thumbsup:

[avandonk]

ImagesPlus is your best friend for this! If PS can split into R,G & B tifs fine.

Bert

[dugnsuz]

Cheers
Doug

[multiweb]

Hi Bert, huge field. Great shot. :thumbsup: Why are the stars square? Is it undersampling or something else?

[Omaroo]

Great effort Bert. Not too sure about the over-abundance of stars as it makes the image really over-busy IMHO, but a nice wide field nontheless. :thumbsup:

[strongmanmike]

Wha tha??? :eyepop:Chris, Chris, Chris...that's what's in the sky ma man! :lol:

"Too many stars"....now I've heard everything :rofl:Carl Sagan would have you for heresy.

Mike

[strongmanmike]

Pretty awesome field that Bert!

I like the framing, has interest..need to remove a few stars though ;) :lol:

Be gooder if you could get rid of even more of the gradients.

Nice work

Mike

[gregbradley]

That's an awesome image Bert.

Greg.

[Omaroo]

LOL! Hey Mike - my opinion only - so please shoot me! :lol: When there are too many dots you can't see the nebula!

[avandonk]

Thanks all for the comments. Yes the stars look square if anyone wants it, I will put up the 43MB jpg where they don't (the original tiiff is 235MB). The problem is that the sampling diagonally is at root two the sampling vertically and horizontally. I have never seen this minor criticism with low resolution crappy jpg's that don't do the original image justice. So why start now on a high res image. The Canon 300mm F2.8L has more resolution than even the Canon 5D sensor.It should be interesting what it can do with the Canon 5D mark II. Thanks Greg and Mike as I value both of your opinions.

Omaroo I only shoot 'em I don't make 'em. If you want to see dim nebs you get lots of stars.

I do all my imaging from a light polluted site (home 16k from Melbourne) so gradients are a real pain. I think setting up at a dark sky site would be even more painful.

Bert

[Ian Robinson]

Very nice - bit red though (I believe) .

Stars look very square - ... just saw your post below - an artifact of processing .

I'd love to see the full image - too big to email as an attachment , even compressed by WinZip I expect. Got a link to it ?

Those 300mm F/2.8 Ls are good gear , too pricey for me right now , but I still lust for one or maybe a 200mm F/2 L . I can't justify that kind of expense on one telephoto for the 40D.


Wonder how those Sigma 300mm F/2.8 APOs compare - still outside my bracket though.

Wish it was possible to lease these lenses for a week, fortnight or month (like you can in the states) at reasonable (CHEAP) rates.

[avandonk]

Ian nothing comes near to the Canon super telephoto lenses. You have the proof before you.

Bert

[Omaroo]

Thanks for advising me of that. I wasn't aware. ;)

After re-visiting the higher resolution versions I think it is, indeed, the squareness of the stars that throw me. They probably look more prominent on my screen, to me, at full scale because of this. If I had an incredibly large monitor that let me stand right back a few feet to view the image at full size and 72dpi, I'm sure it'd look utterly fantastic. Seeing small segments and panning around on a desktop screen doesn't do your supply of a full resolution image justice.

I've also not seen many super-wide field renditions of the Eta Carinae area - most images I've seen to date are of the central core - and the Homunculus/Keyhole in particular. Around these features the starfield has always appeared less dense in the shots I remember, that's all. Is this because of the long focal lengths and highr f/ratios used to capture this local area full frame are not letting a rich star field develop?

I can see now after looking at several other images I've Googled this morning that the starfield is in fact incredibly dense in several widefield presentations. It never clicked with me that the Eta Carinae region in widefield was so dim. So with that go my apologies.

Mike - thanks for making me appear a goose. Go eat bananas. :lol:

[multiweb]

Wasn't a criticism at all Bert. :) I don't understand how downsampling a high resolution image would make these kind of star shapes. I compress large frames to JPEG format in Photoshop all the time prior to uploading but I never had a problem with the resampling of the pictures. Do you use Photoshop to resize?

[rmcpb]

Amazing image of one of the show piece areas of the sky.

[bojan]

Bert, when did you take the central part of your image?
Just in case, I marked the area where Nova Carinae appeared.. there is something very faint there on your image (sorry, I had no patience to wait for less compressed image, even with this 5MB I had more than 10 download interruptions before it came through).
Have a look.
Regards,
Bojan

[avandonk]

Bojan it was taken months ago. I would have to check.

Bert

[bojan]

No worries..
It is a stunning image, BTW.. sorry for not mentioning this before, I am sort of in "nova detective mood" at the moment :-)

[avandonk]

Bojan here is your image superimposed on a crop with Registar. If you would like your whole image done this way just PM me and send me you email address.

Bert

[avandonk]

Here is a version where the stars are not as square! If you wonder how I did it lookup Niquist Theorem.

15MB image
http://avandonkbl.bigblog.com.au/data/2/13839/image/mosnew1329220081211143753.jpg


Bert

[bojan]

:-)
Bandwidth rules...

[avandonk]

It took me a while to figure it out. It is obvious when you think about it. In a typical digital image the image is sampled at the pixel spacing p vertically and horizontally. At 45 degrees the image is sampled at 1.414p or root two times p. So round stars will tend to look square if the sampling per star is low. Especially after stacking to enhance signal to noise.

How to fix this? My Canon 5DH images are 4358x2912 pixels and each pixel is 8.2 micron which is larger than most DSLR's. Rotate the image by 45 degrees CW in Photo Shop. Then REDUCE the size of the image from about 5k pixels wide to 3k pixels wide this has the effect of resampling the original horizontal and vertical to about 1.4p. Then increase the image to 7k across which now resamples the entire image with more pixels per star than the original. Then rotate 45 degrees CCW and again increase the size to 7k across. You now have an image that is sampled the same in any direction. A tiny bit of star size reduction and RL enhancement in Images Plus. At this 7k size the image is a 190MB tiff. When you are happy with levels etc reducing to about the original size gives an image with nice round stars.

I did this to the four images that made up the mosaic. And then put them together with Registar.

It is obvious when you think about it. The result proves it. I am sure a better protocol could be worked out. I would be interested if any others have had this problem.

Bert

[Peter Ward]

I'd suggest there is more to it. While each photosite is 8.2 micron, they are part of a 16.4 micron Bayer matrix.

De-bayering is never kind to point sources. While one can post -process in a number of ways, if the image is undersampled (eg square stars) no real increase in resolution can be made unless the focal length is increased (ie oversampled)

[avandonk]

You are correct Peter but this does not and cannot increase 'resolution'. All I have done is make the sampling quasi isotropic and at about the level of the diagonal sampling. Bayer interpolation and anti aliasing filters make the evaluation of resolution from first principles vague.

The last image if you are viewing it at 100% on a typical LCD monitor is about 1.4 meters wide.

Bert

[Peter Ward]

Depends whether we are talking about resolution on the sky. With a 300mm lens and 16 odd micron Bayer matrix, then the system will resolve around 11 arc sec per pixel. eg you could not split say a 4 arc second double.

Going to 3000mm you will have 1.1 arc seconds of sky on the same Bayer group. Our double star from before is within easy reach.

Peter

[bojan]

Do you really have to cover the whole Byer group?
I think individual pixel sizes are the resolution limitation for luminance. The whole Byer group is resolution limit for chrominance.. not much different from analogue TV stuff (where BW for chrominance signal is 1/3 of that of luminance). De-Byering process should take care of that...

[Hagar]

Hi Bert Your new method sounds interesting but it also sounds somewhat complex and could lay traps for the inexperienced . namely me.
Nicely processed though with a nice overall aspect.

[Barrykgerdes]

Hi

I don't care what anyone else says.

Bert's two great High resolution images, LMC and Eta-Carina suited my use better than anything I have ever had before.

I was looking for two textures to add to the display in Stellarium that could blend in perfectly with the star background. I had previously tried many smaller angle pictures but they always looked like added photos and not part of the sky scene.

I did have to process them into squares at a size of 3072 pixels square to get the maximum detail I then tried the resulting picture in Stellarium but alas it was too much for my computer. I reduced the size to 2048 x 2048 this did load but the field was too big to accurately position it. I could get the corners right but the middle was out too far.

I then had the idea of converting back to a mosaic so I carefully cut the picture into four squares 1024 x 1024. This was a great improvement. I think I could have got better results if I made a mosaic of nine from the original 3072 x 3072 pic. I will do this sometime.

I eventually reduced the sizes to 512 x 512 because loading all textures into the program started to take up too much memory but they still look great and the lower resolution is not noticeable except in deep zooms into the display.

The roundness of stars at this resolution is not a problem as when zooming in on a display the stars in the program sit very nicely on the background of the texture.

I have a couple of screen shots on my Stellarium web site
www.geocities.com/wendygblyde (http://www.geocities.com/wendygblyde) These are from a screen resolution of 1680 x 1050. This should give some idea as to how good textures can be blended into stellarium.

Barry

[Peter Ward]

I don't think so. While there are many De-bayering routines, some better than others, and they all rely upon extrapolating or interpolating both the brightness and colour from G-R-B-G matrix.

With smooth gradient "daylight" type images the process works very well, but with point sources, less so.

http://www.atscope.com.au/BRO/tutorials/astro_newt.html

Is an example of sections two images taken with similar pixel size, aperture and focal length telescopes (ie similarly oversampled). If you look closely at the small and closely spaced stars particularly at top right, you can see the loss of resolution due the Bayer matrix.

Hardly severe, but present just the same.

[bojan]

Peter,
Thanks for this clarification :-)
Yes, the effect is there but as far as I am concerned, negligible :-)
Actually, I have noticed that on hot pixels. One would expect only one pixel light up, which is true, but there are also surrounding pixels visible, only at much lower intensity.
However, as I said, not important at all for what I am trying to do.