Noise reduction techniques for Canon G9

[gordonsbuck]

A number of us have the Canon G9 and are attempting to deal with the noise at the higher ISOs. I've written about one set of experiments on my blog, http://lightdescription.blogspot.com, but there remains much to be learned. Why not make a joint effort?

Please don't jump in with "... sensor too small", "camera no good", "get the exposure correct", "get a Fuji", etc., etc. My intent is to generate dialog from meaningful experiments and to share techniques and best practices for processing RAW files specifically from the Canon G9.

Certainly there are more than one set of answers but, at the same time, some of this processing can be done with Photoshop Actions (preferably while I'm sleeping).

At this point, summarizing from my blog, I think ISO400 from G9 RAW can be very useable for up to 8x10 prints. ISO 800 is iffy; 1600 is probably not useable for 8x10 prints -- but I am hoping!

I use Neat Image. I've made the Neat Image profiles from the Neat Image calibration prints and noted that the profiles are different according to how the RAW image is processed in ACR. The calibration print profiles are also different from the auto-profile for a specific RAW image file. (I've not posted my Neat Images files because ACR for the G9 RAW is still a Beta.)

It seems to me that noise reduction is best done immediately after opening the RAW file. I've found that, as expected, the Neat Image profile generated from the Neat Image calibration prints varies according to how the RAW image is processed in ACR. Most significantly, assuming "correct" exposure, a Neat Image profile varies with ACR noise reduction, if any, and ACR sharpening.

It's easy to say "don't do ACR noise reduction or sharpening" and that may be the conclusion when all is said and done. However, my sense of it all is that there must be some optimum with respect to ACR noise reduction and sharpening. That is, the ACR noise reduction is probably not based on exactly the same algorithm and settings as, say, Neat Image. Therefore, there may be some optimum combination. Same reasoning with respect to ACR sharpening and PK Sharpener (my choice).

At this point in time, my post processing of G9 ISO 800 RAW files uses

ACR Detail page settings: 50, 0.5, 20, 80, 0, 100 (slight change from my blog)
Neat Image: noise profile from Neat Image ISO calibration print, half strength
PK Sharpen: capture sharpen, high resolution camera, medium or narrow to taste

It now seems to me that the luminence setting in ACR tends to reduce detail (not apparent in Neat Image calibration prints) and generates "plastic blotches" (to use a technical term). Therefore, I've decided to set luminence noise reduction in ACR at nearly zero.

So, how about some other techniques and settings?

[NITeam]

I use Neat Image. I've made the Neat Image profiles from the Neat Image calibration prints and noted that the profiles are different according to how the RAW image is processed in ACR.

This is to be expected. Different post-processing settings (before an image arrives to NI) produce versions of the same image with different noise properties. So, profiles built by NI from such versions also reflect that difference.

I've not posted my Neat Images files because ACR for the G9 RAW is still a Beta.

We have recently published a series of profiles for a similar camera, G7 (with RAW enabled by skillful hands :-), and it seems ACR 4.2 is already able to convert such RAWs. In any case, accurately built profiles are always welcome in the library, even if they are built for beta (they could be updated later when the release comes out).

It seems to me that noise reduction is best done immediately after opening the RAW file.

Actually that is what we recommend to Neat Image users all the time. Any post-processing that is done before NI can potentially do two things to an image: (a) degrade some of the details so that it would be more difficult for NI to detect and preserve them later on, and (b) aggravate noise making it more difficult to reduce and find details in it.
If you consider another noise filter like the one in ACR, I believe it doesn't aggravate noise, but I it may still degrade some useful details. Lets see it this way: suppose you have two tools, one rough and one fine. Would apply the rough one before the fine one knowing that the rough one may eliminate some details that you want to preserve and that the fine one is designed to preserve? Well, in some cases, it may not be an issue, for example, when you want to achieve maximum smoothness of the final result. So, it depends on the intended use of the image.

However, my sense of it all is that there must be some optimum with respect to ACR noise reduction and sharpening. That is, the ACR noise reduction is probably not based on exactly the same algorithm and settings as, say, Neat Image. Therefore, there may be some optimum combination.

I agree with this reasoning in general (and the noise algorithms are certainly different :-), there should be some optimum. You could prepare a lot of versions of the image processed with different post-processing settings and then evaluate these image versions and find the one that looks best to your eyes (and therefore the best combination of settings). A potential problem here is that such an optimum could be task-specific (depending on the intended use of the image) and also subjective.

At this point in time, my post processing of G9 ISO 800 RAW files uses

ACR Detail page settings: 50, 0.5, 20, 80, 0, 100 (slight change from my blog)
Neat Image: noise profile from Neat Image ISO calibration print, half strength
PK Sharpen: capture sharpen, high resolution camera, medium or narrow to taste

It now seems to me that the luminence setting in ACR tends to reduce detail (not apparent in Neat Image calibration prints) and generates "plastic blotches" (to use a technical term).

That is an example of what I was talking about when discussing rough vs fine tool above.

You may want to check if color details in your images are also affected in the same way by ACR. It is possible that NI will be able to better preserve them.

Therefore, I've decided to set luminence noise reduction in ACR at nearly zero.

So, how about some other techniques and settings?

I also invite all NI users who work with ACR to post their experience with noise reduction by ACR and NI, combined or otherwise. Lets share our experiences in this thread.

Thank you,
Vlad

[gordonsbuck]

Thanks for the quick reply and confirmations.

I don't recommend using the G7 RAW profiles for the G9. The G7 and G9 sensors are physically different. Also, G7 RAW is a "hack" -- not to say that G7 RAW is poor quality, just that it is almost certainly different from the G9.

In manufacturing, the first steps are always roughing out!

[NITeam]

I don't recommend using the G7 RAW profiles for the G9.

I didn't either. :-)

[gordonsbuck]

For the G9 at ISO 800, Thomas Niemann (of PTLens fame) has suggested the following settings in ACR Details page for processing RAW images:

Sharpening
amount=60
radius = 0.8
detail = 30
masking = 0

Noise reduction
luminance = 80
color = 100

I've tested Tom's settings against my previous thinking (and posts) on his images and several of my own images. His settings produce very good results. In fact, I'll have to concede that his recommended settings are the best and simplest that I've found specifically for the G9 at ISO 800. Applying Neat Image afterward either adds no improvement or else degrades the image. Also, although my practice has been to use PK Sharpener Capture Sharpening instead of ACR sharpening, I actually preferred the results of ACR sharpening from the above settings; however, I'll grant that the ACR sharpening might simply be more sharpening than "capture" sharpening.

Previously I thought that luminance noise reduction in ACR was contributing to loss of detail. Instead, it seems that the loss of detail was more related to the masking setting. I now use masking set to 0.

Sometimes Neat Image profiles made from the calibration charts obviously are not the best for a particular image and a significantly better profile can be obtained by manually sampling an appropriate area -- if one is available. In either case, it is very difficult to get an improvement over Tom Niemann's ACR settings by using Neat Image.

I haven't given up on Neat Image and certainly not stopped using PK Sharpen. I hope others will try these settings and variations thereof and pass along their recommendations.

[NITeam]

it is very difficult to get an improvement over Tom Niemann's ACR settings by using Neat Image.

Please see my comments above:

Any post-processing that is done before NI can potentially do two things to an image: (a) degrade some of the details so that it would be more difficult for NI to detect and preserve them later on

I guess this can explain your results. A finer tool cannot help after you apply a rougher one.

It makes sense to compare ACR results with results obtained by NI noise reduction alone, without preceding ACR.

Vlad

[gordonsbuck]

I agree. I made extensive comparisons with and without any ACR sharpening and noise reduction.

I'm not so much challenging as asking for help in finding the best techniques and workflow. It could be that I'm using Neat Image at less than optimum settings; however, I've tried making my own calibration profiles, using an auto profile and various manual target profiles. I've tried full strength, half strength and 30% strength (luminosity) settings.

Evenutally, I hope to find some method that is even better than what I have right now but those ACR Details settings produce the best results on the images that I've tested. Sometimes, not often, Neat Image adds a slight improvement.