Pages: 1 ... 9 10 [11] 12 13 ... 16   Go Down

Author Topic: ZERO NOISE technique  (Read 410174 times)

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
ZERO NOISE technique
« Reply #200 on: March 30, 2008, 08:03:56 pm »

Dear Guillermo,

Quote
Right, but I necessarily have to go to 16-bit integer TIFF in the end.

Photoshop works with RGB TIFFs in floating point quite nicely. Even better, you can use http://www.openexr.com/ as an output format. Native support on NVIDIA cards allows for much faster calculations over bitmaps. There is a Photoshop plugin that allows to open EXR files.
Logged

barryfitzgerald

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 688
ZERO NOISE technique
« Reply #201 on: March 30, 2008, 08:56:15 pm »

Intersting thread, have to say though..noise that is grain like, is not really a problem to me. I fear the meltmaster plastic look too much I am afraid.
Logged

ejmartin

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 575
ZERO NOISE technique
« Reply #202 on: March 30, 2008, 09:12:42 pm »

Quote
You can find the equations for converting among color spaces at

http://brucelindbloom.com/

under the "Math" section.
[a href=\"index.php?act=findpost&pid=185496\"][{POST_SNAPBACK}][/a]

Note that there are matrices M for converting among color spaces, in particular for converting from XYZ to rgb.  Furthermore, it sounds like you want to do a manipulation of the luminosity data in Lab space; the transformation between Lab and XYZ is more complicated than a simple gamma transformation, as you can see from the formulae on the linked site.
Logged
emil

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
ZERO NOISE technique
« Reply #203 on: March 30, 2008, 10:14:21 pm »

Quote
I fear the meltmaster plastic look too much I am afraid.

The plastic look is the result of noise reduction eliminating detail along with noise. Improving the S/N ratio of the capture with actual image data instead of mathematical guesswork will not cause a loss of detail; you get more actual subject texture instead of noise.
Logged

dave unwin

  • Newbie
  • *
  • Offline Offline
  • Posts: 17
ZERO NOISE technique
« Reply #204 on: March 30, 2008, 11:31:03 pm »

Very interesting thread and i'm looking forward to the official release!

As a bit of an aside, i have been very pleased with the Timothy Ames Enfuse plugin for lightroom which can provide a similar noise reduction, although without the control that your program seems to offer.

cheers
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
ZERO NOISE technique
« Reply #205 on: March 31, 2008, 09:46:13 am »

Quote
The plastic look is the result of noise reduction eliminating detail along with noise.

Not only, it is quite often the result of suboptimal colour transforms and reducing colour gamuts beyond reasonable. In some cases it is CFA properties too.
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
ZERO NOISE technique
« Reply #206 on: March 31, 2008, 09:48:55 am »

Quote
Note that there are matrices M for converting among color spaces, in particular for converting from XYZ to rgb.  Furthermore, it sounds like you want to do a manipulation of the luminosity data in Lab space; the transformation between Lab and XYZ is more complicated than a simple gamma transformation, as you can see from the formulae on the linked site.
[{POST_SNAPBACK}][/a]

To convert to high gamma space on a PC one needs a destination space and [a href=\"http://www.argyllcms.com/]http://www.argyllcms.com/[/url]
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2000
    • http://www.guillermoluijk.com
ZERO NOISE technique
« Reply #207 on: March 31, 2008, 09:30:37 pm »

One interesting issue that arised on a Spanish forum: a Canon 30D user experienced these artifacts when blending 4 RAW files with ZN: transitions in the blend switching areas are visible:




ZN calculated wrong relative exposures because DCRAW does not apply a correct saturation point for this camera (I found out is also wrong with 40Ds), so the RAW developments went wrong. DCRAW uses saturation 4095 on Canon 30D while it is 3398. We can see that in the pure RAW histogram of a saturated file:




I have set an option on ZN to set the saturation point which makes use of the wonderful option -S in DCRAW, recently introduced by David Coffin. Now relative exposures are correctly calculated:

Calculating relative exposure...
1.tiff: 1 (+0 EV)
2.tiff: 4,17 (+2,06 EV)
3.tiff: 17,06 (+4,09 EV)
4.tiff: 69,78 (+6,12 EV)
DONE

So final image before tone mapping (same exposure as 1.tiff):




Simple 2-curve tone mapping:




The transitions are now perfect, impossible to distinguish exposure gap.
Sample of SNR improvement with respect to 0EV shot (the only one that preserved the window):



And we can see a respectable real 13 f-stops of DR were captured.




CONCLUSION: saturation point of the camera is very important. For those experiencing this kind of problems (30D, 40D) I will implement a camera combo box in the next version of ZN with proper adjustments to fix this. Also the user will be allowed to manually enter the saturation point.
« Last Edit: March 31, 2008, 09:34:37 pm by GLuijk »
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2000
    • http://www.guillermoluijk.com
ZERO NOISE technique
« Reply #208 on: March 31, 2008, 10:08:31 pm »

A brief step by step screenshot taken from Dpreview (thanks to the signatory in the Kodak SLR forum):

« Last Edit: March 31, 2008, 10:13:44 pm by GLuijk »
Logged

BruceHouston

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 308
ZERO NOISE technique
« Reply #209 on: March 31, 2008, 10:29:50 pm »

Thank you again, Guillermo!

I look forward to the camera combo box version as I have a 40D.

Best regards,
Bruce
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
ZERO NOISE technique
« Reply #210 on: April 01, 2008, 12:21:34 am »

Each camera sample may have its own saturation point. More, they depend on ISO settings sometimes. It is better to test both floor and saturation points of any given camera then to rely on the constants hardwired into a programme.
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2000
    • http://www.guillermoluijk.com
ZERO NOISE technique
« Reply #211 on: April 01, 2008, 04:15:47 am »

Quote
Each camera sample may have its own saturation point. More, they depend on ISO settings sometimes. It is better to test both floor and saturation points of any given camera then to rely on the constants hardwired into a programme.

I agree. The problem is that this is a bit advanced for regular users.
What do you think commercial RAW developers such as ACR or Lightroom do? have a huge table of saturation points for each camera/ISO pair, or just trend to clip highlights with a conservative low saturation point?
« Last Edit: April 01, 2008, 04:16:15 am by GLuijk »
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
ZERO NOISE technique
« Reply #212 on: April 01, 2008, 11:12:43 am »

Quote
I agree. The problem is that this is a bit advanced for regular users.
What do you think commercial RAW developers such as ACR or Lightroom do? have a huge table of saturation points for each camera/ISO pair, or just trend to clip highlights with a conservative low saturation point?
[a href=\"index.php?act=findpost&pid=186036\"][{POST_SNAPBACK}][/a]

Normally converters use both approaches, meaning a table of conservative values

Two shots at each ISO setting, one with a lens cap on, the other - fully blown in each channel allow to make a table. Interesting that points for both green channels are not always the same. Next, camera serial number and a table of floor and saturation points make a nice base for improvement of raw conversion.
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2000
    • http://www.guillermoluijk.com
ZERO NOISE technique
« Reply #213 on: April 01, 2008, 12:23:59 pm »

Quote
Two shots at each ISO setting, one with a lens cap on, the other - fully blown in each channel allow to make a table. Interesting that points for both green channels are not always the same. Next, camera serial number and a table of floor and saturation points make a nice base for improvement of raw conversion.

Yes the first thing I thought of when I discovered the saturation point was not always 2^N-1, was that a very accurate RAW development could 'recover' a good amount of information in those shots that were erroneously taken with too high exposure values. In that way, commercial developers would not always be optimum.

Regarding the black level, I thought all cameras had hidden pixels so it was best let the developer analyse them to find the more precise black point to be substracted. At least DCRAW always worked fine for me in calculating that figure that varies quite a lot depending on exposure conditions.
Some cameras (like Nikon) unfortunately substract that offset in-camera, right?

BTW a friend of mine who has a Fuji S3 Pro has reported several times magenta casts in the highlights when using ACR to develop Super CCD RAW files. Probably the reason is a bit too high saturation point in ACR for the R captors in that camera model.
« Last Edit: April 01, 2008, 12:26:16 pm by GLuijk »
Logged

Plekto

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 551
ZERO NOISE technique
« Reply #214 on: April 02, 2008, 08:54:29 pm »

Quote
Yes the first thing I thought of when I discovered the saturation point was not always 2^N-1, was that a very accurate RAW development could 'recover' a good amount of information in those shots that were erroneously taken with too high exposure values. In that way, commercial developers would not always be optimum.

Regarding the black level, I thought all cameras had hidden pixels so it was best let the developer analyse them to find the more precise black point to be substracted. At least DCRAW always worked fine for me in calculating that figure that varies quite a lot depending on exposure conditions.
Some cameras (like Nikon) unfortunately substract that offset in-camera, right?

BTW a friend of mine who has a Fuji S3 Pro has reported several times magenta casts in the highlights when using ACR to develop Super CCD RAW files. Probably the reason is a bit too high saturation point in ACR for the R captors in that camera model.
[a href=\"index.php?act=findpost&pid=186137\"][{POST_SNAPBACK}][/a]

Wow.  Digital photos that look like old-school film.  I'm impressed at how the largest three problems - dynamic range,  noise, and fringing are gone with your software.  You do need a way to manually enter the saturation point of a camera, since I suspect that each camera is a tiny bit different as well due to manufacturing and optical changes.(perhaps slightly different with each lens, even, though I suspect it's a very small value)

Q: what camera was used?  It's impressive to say the least.

I love the idea of applying this to a scanner as well.

Q2:Could you possibly change it to have "slots" that you put the files into, and then you can select which one is what value? (so it knows what order to properly blend everything if you have lots of pictures)  That way you could have 3 or 4 or even 12 exposures to blend together.(why not, digital "film" is not a factor here - most cards will hold 10-12 raw pictures)

This has honestly made me reconsider whether I should be looking at digital or not.

P.S. could someone who has a film scanner show an example of this as well?  Preferably a Minolta Pro with MF slide film?(probably have to use Silverfast, right?)
Logged

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
ZERO NOISE technique
« Reply #215 on: April 03, 2008, 12:23:04 am »

Quote
Wow.  Digital photos that look like old-school film.  I'm impressed at how the largest three problems - dynamic range,  noise, and fringing are gone with your software.

Welcome to the real world, Neo. Digital has obviously come a long way since you used it last; for most subjects, blending images is not necessary to keep shadow noise acceptably low. The only time you really need it is for high-DR situations like backlit subjects, interiors with brightly-lit windows, etc.

Quote
Q2:Could you possibly change it to have "slots" that you put the files into, and then you can select which one is what value? (so it knows what order to properly blend everything if you have lots of pictures)  That way you could have 3 or 4 or even 12 exposures to blend together.(why not, digital "film" is not a factor here - most cards will hold 10-12 raw pictures)

There is no point in blending more than 3 images. 3 images 3 stops apart are very difficult to blend into a single natural-looking image. There's a limit to how much you can compress DR.  See the "do you hate HDR too" thread for examples and discussion of this. 2 images is more than adequate in the majority of cases, and the more images you have the greater problems you have with alignment. And add a zero or two to your estimate of card capacity; the larger cards available can hold hundreds of RAW frames, possibly over 1000.

Quote
P.S. could someone who has a film scanner show an example of this as well?  Preferably a Minolta Pro with MF slide film?(probably have to use Silverfast, right?)

Film does not lend itself to DR blending as well as digital. Film has a non-linear response curve, which makes calculating the correct blend values accurately much harder. And if the film isn't perfectly flat when exposed and scanned, aligning multiple frames to blend becomes problematic.
« Last Edit: April 03, 2008, 01:04:59 am by Jonathan Wienke »
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2000
    • http://www.guillermoluijk.com
ZERO NOISE technique
« Reply #216 on: April 03, 2008, 05:00:32 am »

Quote
You do need a way to manually enter the saturation point of a camera, since I suspect that each camera is a tiny bit different as well due to manufacturing and optical changes.(perhaps slightly different with each lens, even, though I suspect it's a very small value)
Lens does not affect camera's saturation point. I am planning to introduce a 'Calibrate' button, so just by providing ZN a saturated picture it would calculate the exact saturation point to optimise RAW dvelopment for each particular camera.

As far as I know, they don't seem to differ too much between units of the same model: I have tested saturated RAW files from two 40D's and both saturated exactly at the same level (13000 something).


Quote
Q: what camera was used?  It's impressive to say the least.
If you mean in the sitting room sample, my Canon 350D.


Quote
Q2:Could you possibly change it to have "slots" that you put the files into, and then you can select which one is what value? (so it knows what order to properly blend everything if you have lots of pictures)  That way you could have 3 or 4 or even 12 exposures to blend together.(why not, digital "film" is not a factor here - most cards will hold 10-12 raw pictures)
The present version of the program allows for up to 10 RAW files (autolimited, I thought more is simply stupid) and they don't need to be ordered. The program will order them by exposure level and calculate (not read the EXIF) the relative exposure between each pair of images. If the program allowed the user to enter the EV differences between the shots, the result would probably be wrong and transitions would become visible due to exposure differences.


Quote
This has honestly made me reconsider whether I should be looking at digital or not.
Digital cameras have become really nice devices Plekto. Their Aquiles heel today is DR and they are continuously improving. With this technique you can avoid the limitations in DR, but with the important limitation that it requires a tripod and a static scene.



Jonathan, I remember you would be interested in a blending with a 16-bit DNG output. Many people have shown a lot of interest for this option. To do that is no problem as long as we know how to build a DNG from scratch, so anyone who knows about the DNG format and would like to make a pure RAW blending tool, just contact me. I think it is not that difficult.

However, in such an approach, it would be a must to have the possibility for anti-ghosting and allow progressive blending in the border areas, since the result will be difficult to correct making use of the original RAW files. I want to try mi anti-ghosting and progressive blending ideas in the next weeks.
« Last Edit: April 03, 2008, 05:08:40 am by GLuijk »
Logged

MichaelEzra

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1134
    • https://www.michaelezra.com
ZERO NOISE technique
« Reply #217 on: April 03, 2008, 10:39:26 am »

Hi Guillermo,

you could probably contact Ed Hamrick (hamrick.com) regarding DNG output. His VueScan software has that capability, as well as blending of 2 exposres from a scanner input, but not from individually shot raw files. There is probably room for cooperation;)

Best,
Logged

Jonathan Wienke

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 5829
    • http://visual-vacations.com/
ZERO NOISE technique
« Reply #218 on: April 03, 2008, 10:49:20 am »

Quote
Jonathan, I remember you would be interested in a blending with a 16-bit DNG output. Many people have shown a lot of interest for this option. To do that is no problem as long as we know how to build a DNG from scratch, so anyone who knows about the DNG format and would like to make a pure RAW blending tool, just contact me. I think it is not that difficult.

The complete DNG format specification can be found at http://www.adobe.com/products/dng/pdfs/dng_spec.pdf. It doesn't look that hard to write a DNG file once you have created all of the 16-bit linear RGB image data. I look forward to a version of Zero Noise that outputs DNG files with great interest.
Logged

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2000
    • http://www.guillermoluijk.com
ZERO NOISE technique
« Reply #219 on: April 03, 2008, 07:11:25 pm »

Just a sample of why I consider important to calculate the relative exposures ourselves, from whithin the code. The sample image (dinning room) was obtained from a {-2,0,2} camera bracketing, so to adjust exposure of the 0 shot to the -2 we would correct exposure by -2EV.

However ZN calculated less exposure correction: -1,91 to be exact:

Calculating relative exposure...
3.tiff: 1 (+0 EV)
2.tiff: 3,77 (+1,91 EV)
1.tiff: 15,06 (+3,91 EV)
DONE

Let's find out the difference in the final result when applying a -1,91EV correction, and what would have happened if we decided to apply -2,00EV:




The -2EV reveals the transition area (best seen in the bottom pictures that were processed with a curve to enhance contrast) while the -1,91EV correction yields a much better gradation.

To be honest, I didn't care too much on optimising this calculation; I think it can even be improved to focus on the relative exposure on those image areas where the transitions are to happen.


BR
« Last Edit: April 03, 2008, 07:14:28 pm by GLuijk »
Logged
Pages: 1 ... 9 10 [11] 12 13 ... 16   Go Up