[jlaubza]

HI, I've checked the test images over at Imaging resource very carefully and it is painfully obvious that all the in-camera NR settings for the KX smear away detail in the red channel like a hot knife melts butter.

The image area I looked at was the standard studio set up of swatches of material, coffee cup, beer bottle, Crayola's etc. Look at the red and black patterned swatch to see how Pentax NR loses detail in red areas.

In comparison at same ISOs, the Nikon d90 is a saint but the K7 is just as bad.

I've turned off incamera NR and now do NR exclusively on my PC, generally using Neat Image. The result is better than the camera NR but still very sensitive in the red channel. Comparison of the same shots using KX and D90 images show significantly different curves for the red channel - there is something about the way Pentax handles red colour that makes it susceptible to smearing when using NR.

[UnknownVT]

I've turned off incamera NR and now do NR exclusively on my PC, generally using Neat Image. The result is better than the camera NR but still very sensitive in the red channel. Comparison of the same shots using KX and D90 images show significantly different curves for the red channel - there is something about the way Pentax handles red colour that makes it susceptible to smearing when using NR.

It isn't just NR - JPG generally has a hard time compressing in Red.

So you may have to find the right compression ratio to retain the detail that you so carefully tried to preserve.

I even found a difference between different editor's JPG -
eg: PS Elements 7.0 JPG seems to preserve details in red a bit better than my older PhotoImpact 8.
So on more critical photos I may do all my PP in PhotoImpact 8 save to TIFF
then bring the TIFF into PS Elements 7.0 and save to JPG there -
using the Preview function to make sure the details are still showing......

[jlaubza]

UnknownVT, you make an interesting observation about Jpeg reduction - it is true but most of us will use the highest quality jpeg reduction the camera is capable of. In this way, the Nikon d90 does a much better job than Pentax KX although the reduction ratio is about the same.

My point is that you will still get a better image result by doing NR on your PC whereas in the D90, you can happily use the inbuilt NR and save yourself some effort.

Since the D90 also does the NR act better than the K7 does, one could conclude that this is another reason to cast one's view away from a K7 if you want really good quality images.

I'm a bit off the point here but proponents of the K7 keep punting features instead of dealing with image quality limitations (or strengths, as the case might be).

[Rondec]

My point is that you will still get a better image result by doing NR on your PC whereas in the D90, you can happily use the inbuilt NR and save yourself some effort.

Since the D90 also does the NR act better than the K7 does, one could conclude that this is another reason to cast one's view away from a K7 if you want really good quality images.

I'm a bit off the point here but proponents of the K7 keep punting features instead of dealing with image quality limitations (or strengths, as the case might be).

I disagree. I think people who shoot with the K7 (I am one) enjoy shooting with it enough that they are willing to deal with the limitations. I have said often that I feel like the jpeg engine in particular is weak. If I am shooting high iso with my K7, I use RAW exclusively and get better results, but i agree that I shouldn't have to. At the same time, as I mentioned before, the K7 fits me -- it is a fun camera to use and shoot and I am willing to deal with its weaknesses because of that fact.

[JohnBee]

HI, I've checked the test images over at Imaging resource very carefully and it is painfully obvious that all the in-camera NR settings for the KX smear away detail in the red channel like a hot knife melts butter.

Too true!
I stoppled shooting in JPG years ago and never looked back precisely because of issues like this.
Also... I calibrate many of my setup shots with TTL color calibration cards. You wouldn't believe the difference that makes in areas like the red channel.

[UnknownVT]

I took the ISO3200 sample test images from Imaging-Resource.com
JPGs Default NR=2 (http://www.imaging-resource.com/PRODS/KX/FULLRES/KXhSLI03200_NR2D.HTM) ; NR=0 (http://www.imaging-resource.com/PRODS/KX/FULLRES/KXhSLI03200_NR0.HTM)
and the RAW/DNG with default NR=2 ( KXhSLI03200_NR2D.DNG ) - but since this is RAW one can set NR to anything else other than the As-Shot or Camera Settings.

This is the loss of detail in red the OP was complaining about:


RAW/DNG does do better - even on default NR=2 -
using the As-Shot or Camera Settings
Converted with Pentax DCU 4.11 (SilkyPix)


It is improved slightly if one sets the Random Noise Reduction =0


This is the PS Elements 7.0 using ACR (Adobe Camera RAW 5.6) conversion with slight Luminance Noise Reduction = 9


However without any RAW/DNG manipulation the straight JPG with NR=0 is like this:


Personally I think the JPG NR=0 in this tight crop is as good as any of the RAW/DNG conversions.

My previous take on saving to JPG compression can also destroy details in Red -
here's an example of the "best" of the RAW/DNG conversion
using Pentax DCU 4.11 (SilkyPix) with Random Noise Reduction =0
BUT saved using a very high compression JPG - red details are destroyed:


The ACR converted version
saved with high compression JPG:

[JohnBee]

This is the loss of detail in red the OP was complaining about

The trick here is to shoot in RAW and take control of NR.
Global NR(as shown) is not very effective at retaining detail where it matters, and since the blue and red channels are usually the hardest hit for noise reduction, results like these are a good example of what happens.

[UnknownVT]

The trick here is to shoot in RAW and take control of NR.
Global NR(as shown) is not very effective at retaining detail where it matters, and since the blue and red channels are usually the hardest hit for noise reduction, results like these are a good example of what happens.

Good input - can you please tell us simply how to apply NR to parts of the image in RAW conversion -

I haven't been able to figure out how to do this in Pentax DCU 4.11 (SilkyPix), Adobe Camera RAW or LightRoom 3 Beta2 -
there doesn't seem to be the obvious ability to select a limited/partial area so that I can apply partial NR.

Thanks.

[JohnBee]

I haven't been able to figure out how to do this...

Having seen the new Photoshop CS5 and LR3 beta's. I was truly disappointed to find that Adobe did not include the adjustment brush options to control noise reduction.
And so we are left with the same old methods as earlier. Though at least now the Adobe color noise reduction performance is on parr(if not better in some cases) than most of the better the 3'rd party filters out there so we get to save steps in the workflow.

Coming from either ACR and LR3, the short recipe would be to process multiple copies of the image into Photoshop and mask in the NR areas selectively.

Coming from a 3'rd party plugin such as Noise Ninja, Noiseware, DeFine etc. you can simply brush away(at any given opacity) the NR in select areas in the scene.

[UnknownVT]

Having seen the new Photoshop CS5 and LR3 beta's. I was truly disappointed to find that Adobe did not include the adjustment brush options to control noise reduction.
And so we are left with the same old methods as earlier. Though at least now the Adobe color noise reduction performance is on parr(if not better) with most of the better the 3'rd party filters out there so we get to save steps in the workflow.

Coming from either ACR and LR3, the short recipe would be to process multiple copies of the image into Photoshop and mask in the NR areas selectively.

Coming from a 3'rd party plugin such as Noise Ninja, Noiseware, DeFine etc. you can simply brush away(at any given opacity) the NR in select areas in the scene.

Thanks for that input -

I was concerned I had missed something obvious -
and yes, you're right ACR and LR3 ought to include the ability to select limited areas and/or the ability to use brushes for NR.

Thanks.

[jlaubza]

Red channel noise reduction

Many thanks for posting the test images I used to make my assertions. Clearly Jpeg compression is a culprit - if you had done the same tests with the D90 images you'd see how much better the Nikon firmware is at dealing with noise without losing detail.

What one learns is that Firmware is a hugely important part of image quality and that raw shooting with a Pentax is the best way to go if you take your images seriously, at least until Pentax improves its jpeg engine.

With other brands of camera, jpegs might actually be 'as good as' raw images processed on your PC.

Regarding noise reduction, I tried DxO on K7 images and it was nothing short of miraculous at 3200 iso.

[UnknownVT]

What one learns is that Firmware is a hugely important part of image quality and that raw shooting with a Pentax is the best way to go if you take your images seriously, at least until Pentax improves its jpeg engine.

Actually the Pentax K-x JPG engine is very good - as attested by many reviews:

from: Pentax K-x Review | Digital Camera Resource Page
" There's actually very little difference between the JPEG and retouched RAW images as ISO 3200, which tells me that Pentax's JPEG algorithms are pretty solid. Post-processing in Photoshop does make a more noticeable difference at ISO 6400, so I'd recommend spending the extra time to do that for your super-high sensitivity photos. "

from: Pentax K-x Review: 15. Photographic tests (Noise): Digital Photography Review
" All in all the K-x's JPEG engine is doing a very decent job. Surprisingly at high ISOs the camera is a lot better than its bigger brother, the K-7, and is one of the currently best performing APS-C cameras in low light. "
and from: Pentax K-x Review: 29. Conclusion: Digital Photography Review
" The Pentax JPEG engine is doing a very good job at squeezing all captured detail into the camera's JPEG files and therefore shooting RAW does not produce a great amount of extra detail. While the image quality at base ISO is generally very good, what we were really surprised about (in a good way) is the K-x's performance in low light. Up to very high sensitivities the Pentax output shows a very good balance between noise reduction and the retention of fine detail in JPEGs (raw output is similar to other cameras in its class). The K-x approach: leaning heavily on chroma noise with more lenient handling of luminance noise results in images with grainy, almost film-like noise characteristics, that show very good detail up to the very highest sensitivities. The K-x is surprisingly a lot better better in low light than its bigger brother K-7, and is no doubt one of the currently best performing APS-C cameras in low light."

It is self-evident that at ISO1600 and above the default noise reduction JPG does show a weakness in red detail - but we are looking at 100% pixel level - however that is NO EXCUSE - RAW does do better (I'd hope so at so at about 4x the file-size) BUT so does JPG with NR=0, and that is such an easy camera setting, far quicker than having to process any RAW file.

In fact look at the above crops the NR=0 sample is every bit as good as any of the converted RAW samples in that limited area -
and that was straight out of the camera, no tedious mucking about with any processors.

Of course RAW does have many advantages -
and its rote is so oft repeated (ad nauseam)
but other than blindly using the default NR in JPG -
this is not necessarily such a clear cut one.

The OP and the title of this thread basically has got it right -
if red detail is important then turn NR Off.

For most practical purposes pixel peeping doesn't do us much service - I do NOT disagree that the K-x JPG default NR does have a severe weakness in Red details - but when shrunk or printed to typical (smaller) sizes - that area becomes hardly noticeable:

Best RAW/DNG processed file with NR=0


JPG default NR=2 (with the BAD red problem)


kind of hard to tell any difference.....

[Wheatfield]

Since the D90 also does the NR act better than the K7 does, one could conclude that this is another reason to cast one's view away from a K7 if you want really good quality images.

I'm a bit off the point here but proponents of the K7 keep punting features instead of dealing with image quality limitations (or strengths, as the case might be).

Any camera built today will provide really high quality images. The tests to destruction that you cite do not often happen in the real world, and when they do, are not likely to ruin a print.
Something that you are not considering in your dissertation is that the most important feature of a camera system is the lenses.
In this regard, Pentax is superior, though YMMV on that.

Go out and take some pictures rather than looking at internet tests.

[glanglois]

Actually the Pentax K-x JPG engine is very good - as attested by many reviews:


kind of hard to tell any difference.....

I think I missed a step: the jpg with NR looks a great deal better on my monitor at the 4 X 6 size - much crisper, livelier, etc.

More to the point, the Fiddler's Elbow reminds me that I'm looking forward to re-acquainting myself with Wychwood's lineup in the next few months. I can get only Hobgoblin (though "only" is the wrong word) around here. Much rather take my time over it in a pub than bring it home from our version of Tesco's.

[UnknownVT]

I think I missed a step: the jpg with NR looks a great deal better on my monitor at the 4 X 6 size - much crisper, livelier, etc.

The default JPG may have some sharpening that I did not apply to the RAW/DNG?
Now that you mention it I can see that as well - I was talking about not much difference in the red patch area - where all the discussion of problems stem from.