ZERO NOISE technique

[Diapositivo]

GLuijk,
In what way would you say your method is better than standard blending procedures as outlined in this Luminous Landscape tutorial?

http://www.luminous-landscape.com/tutorial...-blending.shtml
[a href=\"index.php?act=findpost&pid=124781\"][{POST_SNAPBACK}][/a]

The blending tecnique(s) described in the mentioned articles are some kind of substitute for "Curves" when the scenes has a contrast that streches beyond the dynamic range and curves don't work. The result is always to lighten up the shadows or dodging the highlight in order to recover detail and legibility where it would be lost.

GLuijk tecnique can be used to recover details in the higlights but can be used also with a totally different goal: the lowlights are not made brighter, but less noisy. The picture has the same exposure but looks as if was taken with an "ideal" camera, with almost no noise.

I find this is new, extremely interesting, and absolutely to be encouraged.

Also I find that if the tecnique works with scanned film, this would give film a renewed interest, because only if you shoot film the tecnique can be applied also to handheld shots, and to shots with flames, leaves moving under the effect of the wind, persons running... in the slide everything is still.

I find this extremely interesting

Fabrizio

[Guillermo Luijk]

GLuijk tecnique can be used to recover details in the higlights but can be used also with a totally different goal: the lowlights are not made brighter, but less noisy. The picture has the same exposure but looks as if was taken with an "ideal" camera, with almost no noise.

It's funny that absolutely ANY blending tool (PS HDR, Photomatix,...) has necessarily to do this at some initial stage: blend the images without altering its original parameters. Unluckily none of them stops there, and all seem to be wanting to fiddle with your image: tone mapping, brightness and contrast adjustments, strange exposure corrections,...

Thanks.

[Diapositivo]

I am new to this forum (and its interface) and when I wrote this comment to an insertion in the first page I did not notice that there where 7 more pages of comment. So I read them all now. Absolutely instructive.

I discovered a version of your program is already born, but only for RAW formats.

I would like to stress that the two exposures with different shutter time can be made only with tripod and in particular circumstances (still life, very broadly speaking).

But for those scanning film, this tecnique can be used always! This would make a program that would handle scanned TIFFs superarchgigauseful!

So may I timidly suggest that either yourself, or somebody with the algorithm, write an "action" for PS or a stand-alone program that do the same with two TIFFs, one normally exposed and the other overexposed.

With my scanner (Nikon LS 5000 ED) I tried to acquire 2 TIFFs: one at normal exposure, and one with +2 EV analogue gain (cannot go further than +2). I then used them with two of the tecniques in the abovementioned LL tutorial, they seem to me perfectly aligned and the result is extended DR in a very clean way (using the second of the three methods).

So, if the scanner is capable of producing 2 TIFFs which are perfectly aligned already with PS CS2 (and easily alignable with PS CS3 as far as I understand) and that can be used for HDR (good news) maybe also your noise reducing tecnique could be used with film scans and with enormous benefits (yes also scanners suffer from lowlights noise). I understand doing the thing in RAW is cleaner and gives better results, but I would give it a try.

(Now that I think about it, actually I can save scans in Nikon native RAW format but don't know what am I supposed to do with them and have always ignored the RAW. Maybe I can pass them to the RAW processing software you wrote your program for)

Cheers
Fabrizio

[Guillermo Luijk]

That's possible. Actually I start from RAW files because I want to start from linear TIFF images, and the only way I found to have linear TIFFs from the beginning (since 99.99999% of TIFF images in this world are gamma corrected), was to develop the linear RAWs myself from my code.

But the concept is general and can be applied to gamma corrected images. The main problem is not actually the gamma (undoing it is trivial Maths). The problem is that TIFF files are usually not just gamma corrected images. They have also been converted to some colour space (means matrix operations that modify all RGB data) and have been adjusted in contrast (means all levels have moved again), brightness and so forth.

No idea what kind of TIFF and parameters can be adjusted in a scanner (never used one), but we would need TIFFs with as few adjustments as possible (apart from gamma). BTW if gamma can be set to 1.0 in the scanner settings (I doubt it) it would be perfect.

[MichaelEzra]

GLuijk,

I have Nikon Collscan 8000 which can generate NEF raw files.
I you would like, I can provide samples at various exposure levels, let me know the specifics.

[Guillermo Luijk]

GLuijk,

I have Nikon Collscan 8000 which can generate NEF raw files.
I you would like, I can provide samples at various exposure levels, let me know the specifics.
[a href=\"index.php?act=findpost&pid=142483\"][{POST_SNAPBACK}][/a]

there is actually no limitation, just try to set the same white balance in all the shots. sorry, in all the scans lol

how much can you increase the exposure level?
if you provide me with 3 NEF scans, the regular one with no blown pixels, and two additional at +3EV and +6EV, we can do a nice test.

Send them through yousendit to: gluijk(at)hotmail.com

PS: choose a difficult scene, high dynamic range for instance, with which you didn't feel happy doing just one scan (poor detail in the shadows, noise,...).

[gingerbaker]

I use an automated luminosity mask blending technique using PS2 which I think has a similar effect on noise.

I use it with several exposures, however, as I use it to create extended dynamic range images, rather than using it just to remove noise in shadow areas.

I'm not sure if it will work with only two exposures, however - Glijik's technique looks a lot more sophisticated behind the scenes.

I'll bet there are folks here who could improve this technique with their "Blend if..." expertise.  

Anyway, here it is for Photoshop:

How to do Automated Luminosity Masking


1a) Open up the two images to blend.  If you have a bunch of the same scene using a tripod, and want a blend using the whole dynamic range, it is better to make multiple blends, each of exposures *not too far apart from each other*

1b)Using  move tool, SHIFT-left click to drag darker image onto lighter image
1c) Minimize the 1st dark image

2) Important - - verify alignment by Layer window ( where you see the layers, layer masks, etc) options list - Use Difference mode.  Zoom in is best, so that you can move image by pixil at a time using arrow keys.  -> be sure to set Layers Mode back to normal when done!

3) ALT-left click on eye icon on background ( lower) layer.  The picture should become lighter.

4) CONTROL-SHIFT-ALT- ~ ( tilde key)   creates luminosity mask. Hit all four of these keys simultaneously.  You should now see “marching ants” delineating the highlights.

5) ALT-click missing eye icon (hit the space where it used to be) on the upper layer.  The picture should now turn very dark. i

6) Go to the main “LAYERS” menu at top of screen, , select “Add layer mask” and “reveal selection”.  (Boom!)

to improve:

7) ALT-click on layer mask icon ( on original layer, to the right.  This will be the upper layer of the two, and all the way tot he right)

Gaussian blur of about 3 pixils - apply. This should sharpen the pic.  To check, zoom in, click ( not alt-click) on the layer thumbnail, to the left of the layer mask thumbnail.

To improve more:

9) again, click  ( not alt-click), on the layer mask thumbnail to activate it. Open up the LEVELS box in the Adjustments section, and  adjust blend using “Levels” adjustments.  Basically slide the middle slider to the right to “blend” the relative contributions of the two pictures.

[Natasa Stojsic]

Hi Guillermo,

Any chance for your Software Release before Christmas?

[span style=\'font-size:11pt;line-height:100%\']Optimist [/span]

[Guillermo Luijk]

Hi Guillermo,

Any chance for your Software Release before Christmas?

[span style=\'font-size:11pt;line-height:100%\']Optimist [/span]
[a href=\"index.php?act=findpost&pid=145507\"][{POST_SNAPBACK}][/a]

I hope so! although I will be on vacation in Namibia from tomorrow for the next 3 weeks.

I was this morning shooting at a typical-for-tourists restaurant in Madrid downtown, those places plenty of disgusting bull heads hanging from the walls. And the routine worked perfect to recover all hair texture in the black bulls (there were two of them) which were full of noise in the least exposed shot, at the same time as light areas were not blown. Worked very well.

[Natasa Stojsic]

I hope so! although I will be on vacation in Namibia from tomorrow for the next 3 weeks.

I was this morning shooting at a typical-for-tourists restaurant in Madrid downtown, those places plenty of disgusting bull heads hanging from the walls. And the routine worked perfect to recover all hair texture in the black bulls (there were two of them) which were full of noise in the least exposed shot, at the same time as light areas were not blown. Worked very well.
[a href=\"index.php?act=findpost&pid=145513\"][{POST_SNAPBACK}][/a]

Glad to hear that, Thanks

[span style=\'font-size:11pt;line-height:100%\']Have a nice trip!!![/span]

[user]

please tell me the requirements for zero noise images

[DarkPenguin]

please tell me the requirements for zero noise images
[{POST_SNAPBACK}][/a]

Read this thread ...

[a href=\"http://luminous-landscape.com/forum/index.php?showtopic=17775]http://luminous-landscape.com/forum/index....showtopic=17775[/url]

[Ray]

The blending tecnique(s) described in the mentioned articles are some kind of substitute for "Curves" when the scenes has a contrast that streches beyond the dynamic range and curves don't work. The result is always to lighten up the shadows or dodging the highlight in order to recover detail and legibility where it would be lost.

GLuijk tecnique can be used to recover details in the higlights but can be used also with a totally different goal: the lowlights are not made brighter, but less noisy. The picture has the same exposure but looks as if was taken with an "ideal" camera, with almost no noise.


[a href=\"index.php?act=findpost&pid=142253\"][{POST_SNAPBACK}][/a]

I don't think this is quite true. Whatever method of image blending is used, the principle for getting less noise in the shadows is always to have an image that is correctly exposed for the shadows, which in a high contrast scene necessarily means an image with blown highlights.

Like-wise, in order to get detail in the highlights (with a contrasty scene) one needs an image correctly exposed for the highlights, which also necessarily means an image which is very noisy in the shadows. This applies to Gluijk's method also. There's no getting away from it.

When blending with the method described in the LL tutorial, one always has a degree of control over the individual layers after the blending procedure is completed, just as one has a choice as to how much EC to apply to each RAW image before converting. The result therefore is not to lighten the shadows, but to blend an image with noise-free shadows and blown highlights, with an image with detailed highlights and noisy shadows. How dark or light those shadows are in the final blend is entirely up to you. If you want them darker, then use the 'levels' control for that particular layer to make them darker. If you want them lighter just to see how much noise is there, you will find that there is very little noise, providing the overexposed shot was at least 3 stops overexposed.

[user]

Read this thread ...

http://luminous-landscape.com/forum/index....showtopic=17775
[a href=\"index.php?act=findpost&pid=151017\"][{POST_SNAPBACK}][/a]

I don't quite understand:
do I need to take the same photo two times? with different camera settings each time? and then somehow merge them?

what about photos that we don't have the time or cannot take two times?

and why no camera manufacturer hasn't implement this technique to happen automaticaly?

[jani]

I don't quite understand:
do I need to take the same photo two times? with different camera settings each time? and then somehow merge them?

Yes.

what about photos that we don't have the time or cannot take two times?

Then you're s**t out of luck with those photos; you just have to make the best of the camera's limitations.

and why no camera manufacturer hasn't implement this technique to happen automaticaly?

At this point in time, it probably is too resource intensive to do in-camera. It may be possible some time in the future.

[user]

At this point in time, it probably is too resource intensive to do in-camera. It may be possible some time in the future.
[a href=\"index.php?act=findpost&pid=151053\"][{POST_SNAPBACK}][/a]

mm what if I set the camera to shoot automaticaly 2 times with the minimum time between shots?

[Ray]

mm what if I set the camera to shoot automaticaly 2 times with the minimum time between shots?
[a href=\"index.php?act=findpost&pid=151058\"][{POST_SNAPBACK}][/a]

Buy yourself Adobe CS3 or better still CS3 Extended. These latest versions of Photoshop have excellent auto-alignment features. Ideally you should use a tripod, but I've found this is not necessary if you can hold the camera reasonably steady for a second or so.

Use the 3 auto-bracketed exposures with a +/- 2 stop interval, which most cameras support, and merge to HDR in Photoshop.

[EricV]

But for those scanning film, this tecnique can be used always! This would make a program that would handle scanned TIFFs superarchgigauseful![{POST_SNAPBACK}][/a]

SilverFast scanning software already does this.
[a href=\"http://www.silverfast.com/highlights/multi-exposure/en.html]http://www.silverfast.com/highlights/multi-exposure/en.html[/url]

[Guillermo Luijk]

I don't think this is quite true. Whatever method of image blending is used, the principle for getting less noise in the shadows is always to have an image that is correctly exposed for the shadows, which in a high contrast scene necessarily means an image with blown highlights.

Like-wise, in order to get detail in the highlights (with a contrasty scene) one needs an image correctly exposed for the highlights, which also necessarily means an image which is very noisy in the shadows. This applies to Gluijk's method also. There's no getting away from it.

When blending with the method described in the LL tutorial, one always has a degree of control over the individual layers after the blending procedure is completed, just as one has a choice as to how much EC to apply to each RAW image before converting. The result therefore is not to lighten the shadows, but to blend an image with noise-free shadows and blown highlights, with an image with detailed highlights and noisy shadows. How dark or light those shadows are in the final blend is entirely up to you. If you want them darker, then use the 'levels' control for that particular layer to make them darker. If you want them lighter just to see how much noise is there, you will find that there is very little noise, providing the overexposed shot was at least 3 stops overexposed.
[{POST_SNAPBACK}][/a]

Well Ray, I have to say what Diapositivo said is right: my routine just generates an image free of noise in the shadows, but with the same exposure, bright, contrast,... and everything as the least exposed shot of the set used. Therefore it is very dark and is the user's choice to choose the best way to lift the shadows where and as much as he wants. I wanted a routine that does not modify the original image's parameters in any way.

On this link you can try this with a real image produced by my program (I only converted it to sRGB from linear and re-scaled it down to a reasonable size, but didn't modify any level or characteristic of it): [a href=\"http://www.guillermoluijk.com/download/11m.tif]FREE OF NOISE HDR IMAGE[/url] (13MB).

You will see it is very dark (it has to be, otherwhise highlights would blow as you stated) but the shadows have such a high tonal richness (for being the result of a severe exposure correction down) that you can easily lift them with a curve and will find no posterization or noise problems. Try it!



by applying severe lifting curves...



The image was produced from 3 shots, 3 f-stops apart, in a scene of 13 f-stops of real dynamic range, totally out of reach for any of our cameras in one single shot (Canons and Nikons reach about 8 f-stops):

and why no camera manufacturer hasn't implement this technique to happen automaticaly?

As far as I know, only one vendor has done this: Fuji on his Fujifilm Super CCD SR sensor, which performs real in-camera HDR. It consists of two separate sensors in one, sharing the same surface. They capture the scene with a relative exposure of 3.6EV (I have calculated this figure which after a good number of tests seems to be a constant parameter), and thanks to this the Fuji S3 Pro and S5 Pro can enhance their DR up to 11 f-stops, about 3 complete f-stops more than any Canon or Nikon around at the moment. The two images are independent from each other, and can be extracted and developed separately from the RAW file.

This article is in Spanish but I think looking at the pictures you can find out how it works: FUJI SUPER CCD. IN-CAMERA HDR

This is a cut of the 2 images contained in a RAW file from the Fuji: the S sensor has the task to get detail in the shadows (having the highlights blown), while the 3.6EV less exposed R sensor takes care of the highlights (having too much noise in the shadows):



By properly blending both images (I did it with my own routine) we get both noise-free shadows and non-blown highlights:

[sniper]

When will your software be available?   Wayne