Is this chroma noise reasonable?

krypticide · 07-16-2009, 12:06 AM

One of the things that irks me with my K20D (which seems to be better on the K-7) is chroma noise at high ISO. Please tell me if I'm expecting too much.

I shot this at ISO 1600. If anything it is slightly overexposed; parts of her face are slightly blown out. However, look at the background; there's quite a bit of blue/purple spots. With Lightroom Color NR at 25, the sharp spots sort of fade out a bit, but they're still really evident.

Am I missing something? I think I exposed as near best as I can tell.

(Hm, noticed that as uploaded, it looks less bad than when I view it in Photoshop or Lightroom. Rather odd. But you can definitely still see it in the full size image.)

100% crop, 0 color NR in Lightroom 2.4 (click for full image):

tigrebleu · 07-16-2009, 12:49 AM

This kind of noise seems a bit annoying to me... was the picture pushed in PP?

I get similar results with my K10D. Pushing the chroma noise reduction slider beyond 25 just won't do any difference in chroma nois levels, but the color accuracy will be affected.

You can always play with the exposure a bit. A minute EV compensation could reduce the visibility of the noise, and it could improve the exposure on the model's face as well.

Another (great) solution is to raise the black levels a bit, from 0 t 5 or 5 to 10, for instance. It should reduce the visibility of the chroma noise, but it will diminish the dynamic range of the picture a bit.

Personally, I use this technique a lot to get deeper shadows and reduced noise visibility. I don't care the DR loss, since it's far from catastrophic. This approach works well with dark backgrounds, both in Lightroom and CaptureOne.

And speaking of CaptureOne, it's more efficient at reducing the chroma noise than Lightroom is, IMHO. I usually can push the chroma NR slider up to 50 with very little impact on the color accuracy.

But CaptureOne has less color control than Lightroom and the basic edition doesn't have any lens correction tools (CA, distortion, etc.) The Pro version is complete with vignetting and corner sharpness correction, but it's more expensive.

You might want to give it a try, though.

Hope it helps.

kasv · 07-16-2009, 02:26 AM

This seems to be quite normal to me and typical for K20D. To what I have seen in this forum as well as the notorious review sites, K20D and K7 are indistingushable in terms of IQ and high ISO noise. Dark midtones often come like this. You may lower their influence by enhancing the contrast or darkening them along with some chroma noise reduction (all PP). Eventually, it all depends on what you want to do with your photo. I doubt this would show up at all when printed. When shown in the web in small size no one would recognize this either. -kai

krypticide · 07-16-2009, 08:30 AM

The image uses default Adobe import settings (5 for black point, brightness 50, and contrast 25 with everything else 0), except that I reduced the default chroma NR from 25 to 0, and I also changed the white balance which was originally too warm. The chroma noise was less noticeable in the "As Shot" white balance.

I wonder if it is possible to reduce chroma noise by shooting at the correct white balance? I mean, this was a RAW file, but I wonder if the camera somehow takes different data depending on measured white balance?

Here's a version with some post processing.

Still, I'm wondering if anyone out there is able to get better chroma noise results than this, without resorting to PP.

**Marc Sabatella** · 07-16-2009, 11:14 AM

Originally Posted by krypticide

The image uses default Adobe import settings (5 for black point, brightness 50, and contrast 25 with everything else 0), except that I reduced the default chroma NR from 25 to 0, and I also changed the white balance which was originally too warm. The chroma noise was less noticeable in the "As Shot" white balance.

I assume you reduced the chroma NR to 0 because you were curious how it would look, not because you thought you were doing yourself any favors in terms of noise? I mean, it should go without saying that you'd get less chroma noise if you reset the chroma noise back to its default (higher) value, wouldn't it?

Anyhow, yes, change of color via WB can affect noise; I see this all the time shooting in strongly colored stage lighting. What happens is that you are altering the relative contributions of each of the pixel types (R, G, or B). If one of those was receiving relatively little light compared to the others (say, the B channel in incandescent light), then it has a relatively low signal-to-noise ratio. Attempting to correct the WB means you are amplifying that B signal more than others - and with it, amplifying its noise. This is something I've learned through experience and thinking through what happens; I can't say that I've seen anyone write up anything really specific on the matter. but I'm pretty sure I'm right about this.

I wonder if it is possible to reduce chroma noise by shooting at the correct white balance? I mean, this was a RAW file, but I wonder if the camera somehow takes different data depending on measured white balance?

It does not. According to folks like GoronBGood who analyzed the RAW data itself, some of the 6MP cameras did to a small extent - they'd perform some processing magic on the sensor data (perhaps before the A/D conversion?) to reduce red channel clipping if you shot under incandescent light with the appropriate WB setting. But it was a small effect, and only on those certain older models.

Still, I'm wondering if anyone out there is able to get better chroma noise results than this, without resorting to PP.

Resorting to? You say that like trying to get good results is a bad thing. Your foreground is well- or even over-exposed, but the background is underexposed, and hence shows noise. Luckily, the foreground is what you care about, so just a small tweak to the levels/curves will hide that shadow noise. And why on earth wouldn't you want to use at least the default amount of NR? That's probably no more than the camera would have done by default had you shot JPEG with the Weak NR setting.

krypticide · 07-16-2009, 11:50 AM

Excellent post, Marc. Thanks for the info.

I'm just trying to figure out if I can get less chroma noise than I already am. While I can increase the color NR, I still don't like how there are blue/purple splotches in the background. However, it seems like I'm not doing anything wrong and I'm simply hitting the limits of the sensor.

So to get a decent image without obvious chroma noise, I essentially end up compressing the dynamic range of the image by increase the black point, etc. Is this why people say dynamic range decreases as ISO increases?

**Marc Sabatella** · 07-16-2009, 12:17 PM

If you use curves, you aren't really changing the dynamic range of anything except those deep shadows, and who cares about those, anyhow? Also, I would think simply applying chroma NR (the thing you turned off!) would eliminate the need for any level curve or level manipulation.

kenyee · 07-16-2009, 02:47 PM

AFAIK, what you see is normal w/ the K20D...that's exactly what Herb Chong on FM complained about (it affects landscape photos quite a bit because you tend to shoot a lot at sunrise/sunset).
If you try it w/ the K7, please post your results. AFAIK, it should be a lot better from the other thread on high iso noise...the noise is grey instead of mottled red...

tigrebleu · 07-16-2009, 06:04 PM

Originally Posted by Marc Sabatella

Anyhow, yes, change of color via WB can affect noise; I see this all the time shooting in strongly colored stage lighting. What happens is that you are altering the relative contributions of each of the pixel types (R, G, or B). If one of those was receiving relatively little light compared to the others (say, the B channel in incandescent light), then it has a relatively low signal-to-noise ratio. Attempting to correct the WB means you are amplifying that B signal more than others - and with it, amplifying its noise. This is something I've learned through experience and thinking through what happens; I can't say that I've seen anyone write up anything really specific on the matter. but I'm pretty sure I'm right about this.

Agreed: I shoot stage performance events and WB selection is very important in order to minimize noise, for the reasons you explained above.

Experience as though me to keep my WB around 4800 K instead of 5600 K (daylight) or 2500-3400 K (incandescent light). This way, the noise in the blue chanel (often the most intrusive one) is kept down thanks to the warmer WB. The price to pay for such a WB is that I often end up with blown out reds, but they're less annoying than bluish noise in vertical pattern. A white balance around 4800 K seems the best compromise when shooting RAW and adjusting WB in PP, at least for me.

SWEngineer · 07-16-2009, 10:28 PM

Originally Posted by Marc Sabatella

... Attempting to correct the WB means you are amplifying that B signal more than others - and with it, amplifying its noise. This is something I've learned through experience and thinking through what happens; I can't say that I've seen anyone write up anything really specific on the matter. but I'm pretty sure I'm right about this.

I am sure you are right about this Marc, but it should be very easy to test: Save several JPEG versions of a file with varied WB. In photoshop or equivalent, inspect the B channel. If it looks noisier (or more accurately, it measures noisier - has larger standard deviation - in a dark, uniformly lit select region) with WB pushed towards blue, you have your proof.

The OP seems to eschew PP, but a couple of fairly simple fixes include: substitute low noise channel (perhaps R) for B channel with a mask to do so only in dark areas. Or run a strong blur filter on the B channel only - again masking for, or only selecting dark areas. The mask / selection does not need to be very accurate in either case.

**Marc Sabatella** · 07-16-2009, 10:50 PM

Originally Posted by SWEngineer

I am sure you are right about this Marc, but it should be very easy to test: Save several JPEG versions of a file with varied WB. In photoshop or equivalent, inspect the B channel. If it looks noisier (or more accurately, it measures noisier - has larger standard deviation - in a dark, uniformly lit select region) with WB pushed towards blue, you have your proof.

Oh, I don't even need to go that far - I see noise appear before my eyes as I adjust the WB slider to correct color. The question isn't *whether* it's happening; it's whether my explanation of *why* it's happening makes sense.

krypticide · 07-16-2009, 10:57 PM

Okay, well, at least it doesn't sound like the camera is bad or I'm doing anything inherently wrong. Looks like for shots that really matter, I'll just have to spend more time tweaking them.

I think this is the only thing that bugs me on (Pentax?) high ISO: not so much the grain-niness but the strong chroma noise under tungsten white balance. Chroma NR gets rid of the sharp dots, but you can still see splotchy blue/purple spots.

It looks like the K7 may have improved on the chroma noise, perhaps with help from better auto white balance which helps with appropriate metering, so I'm hoping that is the case.

wll · 07-17-2009, 03:17 AM

Originally Posted by krypticide

Okay, well, at least it doesn't sound like the camera is bad or I'm doing anything inherently wrong. Looks like for shots that really matter, I'll just have to spend more time tweaking them.

I think this is the only thing that bugs me on (Pentax?) high ISO: not so much the grain-niness but the strong chroma noise under tungsten white balance. Chroma NR gets rid of the sharp dots, but you can still see splotchy blue/purple spots.

It looks like the K7 may have improved on the chroma noise, perhaps with help from better auto white balance which helps with appropriate metering, so I'm hoping that is the case.

From what I've seen of K-7 high ISO images the chroma noise is is quite a bit improved compared to the K20D. One of the things they did right on the new camera IMHO.

wll

**Marc Sabatella** · 07-17-2009, 01:30 PM

Originally Posted by krypticide

I think this is the only thing that bugs me on (Pentax?) high ISO: not so much the grain-niness but the strong chroma noise under tungsten white balance. Chroma NR gets rid of the sharp dots, but you can still see splotchy blue/purple spots.

I guess that depends on how much chroma NR you apply. I find with my current software (ACDSee Pro 3 beta), I can crank the chroma NR to the max and it gets rid of most of it, and then applying just a a touch of luminance NR gets rid of most of the rest, and then I can always push my darks a bit darker and get rid most of what's left, to the point where it's basically a non-issue. That's with the K200, but I find the same basic approach works well with the K20D RAW files I've looked at.

Sure, the K-7 might (emphasis on might) do better out the box, but it strikes me as a $1000 solution to a 99 cent problem, given how easy it is to deal with in PP.