K20D vertical banding

On the weekend I finally found time to take a stroll in the autumn landscape. At the shore I took this photo (with the Sigma 105 macro, EXIF intact):
[IMG]http://i156.photobucket.com/albums/t5/Prieni/_RDP4340pu.jpg[/IMG]

8 seconds later I took a second shot with same settings:
[IMG]http://i156.photobucket.com/albums/t5/Prieni/_RDP4341pu.jpg[/IMG]

Back home I started converting the day's bounty with SilkyPix. One step of the routine is to correct CA and for that I usually zoom to 400% in one of the corners. When working on the first shot I noticed regular vertical lines repeating, one line with lighter pixels, one line with darker pixels etc. This pattern filled the whole image. Here's a 100% crop of the lower left corner of the first shot:
[IMG]http://i156.photobucket.com/albums/t5/Prieni/_RDP4340crop.jpg[/IMG]

The same for the latter shot:
[IMG]http://i156.photobucket.com/albums/t5/Prieni/_RDP4341crop.jpg[/IMG]

So two questions:
1. What the heck is going on there?
2. Should I put this thread up over on depressing review and stop the time until the first one bites?

Ok, question 2 is not that serious.
I'm not overly concerned as it hasn't happened for the last 4000+ shots. But something similar happened when the camera was new (see thread here)

Check the EXIF data with PhotoMe and see if the SR had stabilised. It looks to me as if the SR may not have been stabilised and the banding is not quite as you say.

EG: Have a look at the top left edge of the triangular shaped stone in the bottom of the pic that shows the problem magnified. It's as if maybe the sensor was still moving when the shot was taken.

Just an idea but might it be the temperture... dark current?
Only guesing now *sighs* it really shouldn't happen with such a camera.

Does anyone have an idea it's the sensor or the in camera processing that cause this?

The GX-20 has the same sensor but Samsung uses there own software if I'm correct, has these models have the same problem?

Stefan.

-edit-
*Quote* Banding noise is highly camera-dependent, and is noise which is introduced by the camera when it reads data from the digital sensor. Banding noise is most visible at high ISO speeds and in the shadows, or when an image has been excessively brightened. Banding noise can also increase for certain white balances, depending on camera model.

So it's an hardware thing right

Peter,

looking at the EXIF that seems to be the (only) difference between the two shots. When taking the first shot SR had not stabilized after 0.42 s; on the second one I gave the camera 1.2 s and stabilization was effective. So this might be the reason, although I don't quite understand why that would cause the banding.

Now you're hurting me, not trusting my description of the banding
It is possible that I didn't describe the banding correctly, just to make sure that we talk about the same thing I have added a 200 percent crop (upscaled in Photoshop using 'bicubic'):
[IMG]http://i156.photobucket.com/albums/t5/Prieni/_RDP4340crop200.jpg[/IMG]
I think that this shows exactly what I described (at least it does look exactly as what I wanted to convey in words :wink.

Stefan: The camera temperature was 16 degC in both cases (and I doubt that it would change very much in 8 seconds anyway).

So for the moment I will blame it on the SR not being ready. Have to dig out other shots where SR was not ready and see if that showed the same thing...

Cheers,
Prieni

I don't know if it's relevant, but there is some evidence of flare, so maybe it's a diffraction effect because of the light and the orientation of the microlenses over the sensor.

Prieni, I probably didn't explain exactly what I meant either. In some parts of your image the banding is exactly as you described, alternate light and dark lines. The part of the pic that made me question that description slightly is the bit I pointed out where that stone seems to have parts of the darker area above it mixed in along its top left edge giving it a castellated look. It's that that made me suspect the SR maybe wasn't stabilized, which seems to have been the case. Whether that is the only cause of the banding remains to be seen.

I don't know why it would cause the banding either except that the SR system might generate this kind of noise if it is still trying to stabilize when the image is exposed. Certainly the castellated effect might be due to the sensor being in motion at the time the shutter is released.

I'm not sufficiently knowledgeable about the way images are taken from the sensor. If it works in a similar way to a TV display which is interlaced (all odd lines followed by all even lines), by the time all odd lines are captured, the sensor could have moved before the even lines are captured. If it's not stabilized, it will not have 'locked on' to the image and could end up anywhere.

I think Pentax should issue a firmware update for all SR equipped bodies. If SR is on, give us the option to delay shutter release until SR has stabilized. They have done it with AF, so why not SR?

Keep us informed of your findings.

*Quoted* from here link
In short, banding noise = read-out noise. I am no expert on CCD and CMOS sensors, but here's a brief take on this:

Unlike other types of image noise, this is primarily induced by the way the pixels are read as opposed to from thermal noise or statistical variation in the number of photons reaching each site. It results from the pattern/way at which the pixels are read (multiple simultaneous read-outs), combined with performing this read-out at a high speed.

Many newer cameras have problems with banding noise primarily because the pixel information is read in parallel (by several read-out channels) in order to increase how fast they can access the pixel values--and therefore how high of a frame rate the camera can achieve.

While I know this is not your camera, the Canon 20D, for example, has a 4-channel read-out system which allowed it to up the frame rate, but at the expense of increased banding noise in the shadows.
*end*

I don't know how the pentax SR works but it might be that it uses more power to stabilze it's self or that an exra device is used for that? but my idea is that maybe one of the powerlines for stabilizing the SR system might run to close to one of the channels of an read-out system. So when power runs over it, it might cause interference.

I'm just speculating here but it does connect the banding noise with the SR. There is also no prove that the SR system is at fault.
I just don't think that the senor can move so neatly to a side that it leaves perfect vertical lines.

Can you see it except when pixel-peeping?

G

Sorry, but I don't really see what the problem is. It is only visible as far as I can see when you go up to at least 200%. I can't see it on the 100% crop and certainly not on the original photos you've posted here. Now maybe my eyes aren't the best, but in all honesty does it really matter - are you ever going to use these photos, or any other for that matter at such a size that it will be visible?
What is noticeable in your 100% crops is how much sharper the second photo is.

Thank you all for ideas and suggestions.

John,
the scene is very similar for both shots and the flare is there in both shots. So why should it influence one but not the other?

Peter, Stefan,
as far as I am aware the image is captured first (in full) and after the shutter is closed the readout starts. So I don't think this banding is from movement of the sensor while capturing the image (though I expect when SR hasn't locked on then the sensor IS moving during capture).
The suggestion for a firmware update does sound reasonable.

Even with a parallel readout system (and judging by the frame rate it couldn't be a very effective system in the K20 if it has one) I wouldn't understand why this difference should only occur when SR hasn't locked on.

George, Gwyn,
you can see it at 100% if you try hard (i.e. you better know before what to watch out for). So this is a 'no' on George's question (if only just).

And this is why I am not overly concerned. Having said that, if I would like to print this beautiful photo at 50 x 75 cm I guess that it would matter. As it is not such a great shot (and I have the second one without this 'feature') it doesn't matter at all.

I would like to find out what the reason for this banding could be to avoid it in future (it could be the 'once in a lifetime' shot that would be ruined by it (or by some hundreds of other things...)).

If I can pin the reason down I'll let you know.

Cheers,
Prieni

Thanks Prieni,
It's hard to disribe what I mean that i don't think you got me yet but with an picture it works better to word myself.



taken from here link
I know it's Nikon but they had the picture I needed.

As you can see the information from the pixels goes to the Column amplifers and each amplifier set his information on a Signal lane. So amplifier 1 set it's info on line 1, 2 on 2, 3 on 3 ~ 12 on 12, 13 on 1, 14 on 2, etc. so each pixel row is devided over 12 lanes. If one of those lanes get an interferance of somesort the information gets distorted but since this happends only to one lane (for example lane 1) only pixel row 1, 13, 25, 37, 49, 61, 73, etc. gets affected. that's what banding noise is, noise over just one of such lanes.
It can also be an read-out error off such lane by the prime sensor, I don't know but since the only difrence we can find now is the SR system that might be the cause of the disorder. My thought was that it might use extra power or an extra system to stabilize its self and that that might cause an interference on a lane but that's just pure speculation.

Hopefully I'm clearer now, it's so hard to explain somthing in a language not your own.
Stefan.

Going back to the OP - Silkypix has a control for banding noise reduction. It's termed "Geometric NR"
I've seen a LOT worse on other cameras
Matt

It's extremely obvious on the 100% crop with my eyes/monitor combination.

I bet it would be too, on a decent sized print.

Stefan, tell me about it
I did understand what you meant from your earlier post. To get consistent banding over the whole image, though, the amplifier of one (or more) lanes has to receive an interference that consistently affects the amplification. While that is not impossible (I don't know much about the innards of the camera electronics) I would think it is quite unlikely.

Thanks Matt, I wasn't aware that the "Geometric NR" would help with such problems. I certainly will give it a try.
And yes, I don't think it is serious either. If I wouldn't do CA correction at 400% as standard I wouldn't have noticed at all. And when printing I would have blamed my ageing ink jet printer...

Prieni

Okay then it's all good Prieni
but still you need to look at the lanes those are the source of the problem directly or indirectly, atleast without them you wouldn't have had it.

Since I don't have Silkypix, I've tried some stuff with Photoshop on your crop and Filter>Noise>Madian seems to do the trick as well with a radius of 2pixels. I also aplied a copy of the with ontop with a lighten layermode at an opacity of 70% to get some detials back, you can play of course with that to get the right balance.
I've the result here if you want to see it. I really want to compare it with Silkypix, it seems an interesting program but never thought of buying it because I've Photoshop CS3.