In Camera Noise Reduction question

[Schewe]

This has been done numerous times on the Adobe ACR user forum.

I was part of that discussion and I never saw any proof (worth mentioning)...I saw a lot of hand wringing...but after Camera Raw 4.3, the only complaints were from people who didn't know what they were looking at (and what it means).

Is there room for improvement? Sure, you betcha...take the Camera Raw interpolation algorithm for example. Enough evidence of ringing artifacts was proven that the Camera Raw team worked to fix it (thanks Eric) and the 5.2 update proves that if you can prove something, they will work on things. But proof needs to be something other than raw processor A at default and raw processor B at default, crop to 100% and post it on a web site. Ya know?

[barryfitzgerald]

Jeff, the point is not 100% crops, but the quality of my high ISO prints are not as good with ACR, full stop. The only way to show this, aside from dropping around, is for me to show pics here. The other poster also did this, that was with studio shots too, I could certainly see the difference, and in print, bar tiny sizes..

Yes it has been talked about before, but nobody seems to want to do anything about it. People were freaking out when sony were playing about with raw  noise reduction in camera, ok this is not as severe as this. But surely the point of raw, is to do your own processing? And that includes noise reduction.

I want to be able to get the best possible quality for high ISO, yes I expect noise, and I expect to be able to deal with it myself. ACR isn't giving me that choice.

[Schewe]

Yes it has been talked about before, but nobody seems to want to do anything about it. People were freaking out when sony were playing about with raw  noise reduction in camera, ok this is not as severe as this. But surely the point of raw, is to do your own processing? And that includes noise reduction.

Exactly what do you think raw is? The very act of the camera's analog to digital converter "processes" images. Even opening native raw files in Camera Raw essentially converts them to DNG so Camera Raw can read them. Then, you see the result of the baseline demosiacing, noise reduction and some other stuff like color interpretation and default tone curves...and that all is happening before you get to see the preview in Camera Raw. So, you tell me, exactly what you think raw should be?

There are utilities you can use to parse the raw file without doing the stuff Camera Raw does. If you want really raw files, why even bother convert to a color space...keep it linear. Or, better yet, use Rawanalyze (written by a fellow LuLa member) which doesn't even do demosiacing...

The reason your "moaning" hasn't gotten much traction is, well, it's pretty much wrong. Sorry, but there it is in a nutshell. Again, if you view your image at 1:1 or above, you're looking at SciFi stuff that ain't real. What you think is micro-detail is simply unusable noise. Yes, there is a very fine line between noise and micro-detail...in order to step on noise, some micro-detail also gets stepped on. But it's stuff that isn't really usable detail and it's of a size and scale that it would never print.

So, you need to move on beyond this if you want to move the Camera Raw team. I know for a fact that each and every engineer has one singular goal in mind for Camera Raw–o do the best possible conversion from raw to encoded image file. They welcome and encourage feedback and direction (properly given without a lot of 'tude) from everybody but there's a kinda minimum level to which they pay attention to and that's the REAL and legitimate aspects of image quality that Camera Raw can do something about. There is talk about way of improving image quality on a variety of levels...the update to the resample code in 5.2 is a good example of if you prove there is a problem and there's a "better way" to do what Camera Raw is doing, the engineers will move heaven and earth to make that happen.

But you really, really need to know what you are talking about and be able to prove that the "current way" that Camera Raw works is leaving image quality on the table...and the Camera Raw noise reduction simply is not the "really bad problem" some people make it out to be–not for real life situations and uses of digital raw capture.

[Ben Rubinstein]

Given that if I remember correctly Schewe was an advocate of ACR 4.1 and it's horrendously awful over smoothing at higher iso's, I doubt you will find much objectivity from that bullying Adobe Yes man.

[Schewe]

Given that if I remember correctly Schewe was an advocate of ACR 4.1 and it's horrendously awful over smoothing at higher iso's, I doubt you will find much objectivity from that bullying Adobe Yes man.

Actually, you got that part wrong...on the Camera Raw forums, I downloaded the files a fellow provided that were shot with a Canon D40 and did side by side tests with ACR 3.7 and ACR 4.1 on ISO 800+ images. I was convinced enough that I communicated with Zalman and Thomas and "helped" encourage them to address the issue which resulted in modified noise reduction approaches in first 4.2 and then in 4.3. Since that time, I have yet to find an egregious case of too much noise reduction in the demosiacing component of Camera Raw with the exception of the Sony files where in effect, noise reduction was being applied twice–once in camera and again upon conversion in Camera Raw...pretty sure Camera Raw then cut the noise reduction on Sony A700 files over a certain ISO. So, it would be useful if you knew what you were talking about as well doode...(useful also to know that facts).

[sniper]

If they can cut it out for the sony why not cut it out for all, or better yet give us the option.  Wayne

[joergen geerds]

Noise ain't usable micro-detail...if you are evaluating image detail at 1:1 or larger, you aren't seeing "reality" you're seeing reality magnified x2 up to x4 (or more) and all the stuff people "think" they are seeing isn't usable data...not on prints and not online (unless you routinely crop to a tiny portion of an image or post full rez files on a web site).

I find it humorous when people work themselves into a tizzy about things they think they are seeing when what they are seeing ain't what they're getting. A modern LCD has about a 100 PPI resolution...if you are viewing an image at 1:1 (one screen pixel for one image pixel) you are putting a 2x-4x loupe on all that stuff (even worse if you zoom into 2:1 or higher). And while it's useful to evaluate sharpening settings and noise reduction at those zooms, they aren't showing you real image detail but enlarged detail. Detail at those zooms will NEVER print on paper even if you are outputting at 480PPI on glossy paper. At 480PPI, the image on a display at 1:1 is 4.8x bigger (and if you reduce the size, even to 25% the display is still a low resolution devise).

The only way to evaluate images intended for printing is to print them...and in tests, Camera Raw with "proper" capture sharpening and noise reduction performs really well. So all these posting of 1:1 or higher crops on web sites is really much ado about nothing.

I don't agree with you... looking at pictures at 100% is THE way to evaluate noise, especially when you are planning to print it in huge sizes, like so many people believe they can do. Yes, if you print your source photo at true 300ppi or higher, the noise doesn't affect the image not so much, but if you are planning to print your 12MP at 40x30in or larger, NR comes much more into effect (and not in a good way).

[Schewe]

looking at pictures at 100% is THE way to evaluate noise, especially when you are planning to print it in huge sizes, like so many people believe they can do.

No, it's not...if you are going to print those sizes from small captures, you're gonna have to do a LOT of work upsampling and image processing. I've made a lot of printer up to 44" wide (some to 60" long) and looking at the image at 100% zoom will tell you nothing useful or accurate. 50% or 25% zoom will tell you more but then the problem is that the display is still 1/2 to 1.4 the resolution of the print. The ONLY thing that you can evaluate in terms of print quality is the print. A computer display really doesn't give you accurate information in regards to final print image detail–regardless of the print sizes and in many ways, even worse the larger the print you're making.

[barryfitzgerald]

I would like to point out again, that I am printing.. at small print sizes it won't make any difference.

However, I am seeing other people notice problems too, with ACR for high ISO work.

http://forums.dpreview.com/forums/read.asp...essage=30262093

I am not satisfied that adobe are taking notice of this. It is really simple. Don't do any noise reduction at all. The base level NR is not very effective IMO. And it needs sorting out. No heads in sand on this one. And yes, I am getting better "prints" (larger sizes) by  not using ACR for high ISO work.

If adobe want to tweak their luminance NR slider, fire away. But let's start by not doing any in the ist place.

[Panopeeper]

It is really simple. Don't do any noise reduction at all. The base level NR is not very effective IMO

1. Carrying out basic NR in before/during demosaicing is the right thing to do. Think of following: the noise is more often than not coming from a single channel. Typical case: the noisy sky; it is always caused by the low red channel. If no NR takes place before/during demosaicing, this noise gets carried over in the other channels, the mess becomes worse.

The conditions of doing this and its degree is a different issue.

2. The problem of ACR with the raw images of certain cameras, like the Sony A900 is not in the noise reduction but the interpretation of the raw data.

[madmanchan]

Carrying out basic NR in before/during demosaicing is the right thing to do.

Exactly.

[Max Penson]

The secret for good noise reduction is pleasing grain, where ACR is quite behind. If I had to rank the different solutions in the market in terms of pleasing grain, I'd rank:

Nikon D3/D700/D300 - jpeg - NR low setting
Nikon Capture NX (RAW from D3/D700/D300)
Canon EOS 1D Mark III or newer - JPEG
Canon DPP (RAW from Canon EOS 1D Mark III or newer - filtering holes - use sharpening 2 or less!)
DXO optics (careful - no spatial filtering - don't over sharpen)
ACR
Apple Aperture
Capture ONE