5D Mk2 image artifacts?

[rwheat]

Hi there,

I recently purchased a Canon 5d MkII and have just been doing some lens testing using Norman Koren's excellent test charts and process.

I have been surprised by the following issue and wonder if someone might be able to enlighten me as to the cause - it has me a little concerned.

The attached images represent approx 200% crops of a test target photographed at ISO 100 RAW with in camera noise reduction set to Normal as well as Disabled.  Two images were processed using DPP and two using ACR.  No sharpening or other significant processing was done.

The 200% crops are to make the problem obvious - but it is still visable at 100%.
The issue is the little grid-work of horizontal lines in amongst the vertical bars of the test chart - at around the 70lp/mm mark.

The problem is completely repeatable and varies, as you can see, with NR and RAW processing software.

Any help would be appreciated.  I mainly shoot film - is this issue normal for digital capture with this type of camera?

Thanks,
Richard.

[attachment=11333:DPP_NoiseRedNormal.jpg] DPP NR Normal
[attachment=11332:DPP_Nois...Disabled.jpg]  DPP NR Disabled
[attachment=11331:ACR_NoiseRedNormal.jpg] ACR NR Normal
[attachment=11330:ACR_Nois...Disabled.jpg] ACR NR Disabled

[BernardLanguillier]

Hi there,

I recently purchased a Canon 5d MkII and have just been doing some lens testing using Norman Koren's excellent test charts and process.

I have been surprised by the following issue and wonder if someone might be able to enlighten me as to the cause - it has me a little concerned.

This looks like moire to me.

It is a consequence of the bayer interpolation used to compute RGB colors from the color filtered photosites. Most cameras exhibit moire to some extend when the frequency of a regular patern in the image is close to that of the photosites. It often shows up on fabric,... The AA filter that most DSLR have built on top of their sensor is, among other things, aimed at introducing a slight blurring that is supposed to reduce moire if not prevent it totally.

I don't think it would show much in most real world images, but some raw conversion softwares have built-in moire removal, there are also some photoshop techniques that you should find links to if you google on "moire removal".

Cheers,
Bernard

[francois]

I remember that raw files from Olympus E-1 tended to exhibit "maze" effects with red hues and with some RAW converters.
I must say that I've not seen that kind of maze artifacts on my EOS-1 bodies but I must also admit that I don't shoot test charts.

[francois]

Not helpful but I just read Keith Cooper's review of DxO v5. I noticed that in his article some photos show the maze effects (1Ds3). The article is here  http://www.northlight-images.co.uk/reviews...ptics_pro5.html  (scroll down to the last two photos).

As Bernard mentioned above, it must be a strong case of moiré.

[Panopeeper]

It is a creature of the demosaicing algorythm.

DCRaw offers a selection from different algorythms. Try it out.

[stewarthemley]

It's a LR/ACR thing (and apparently a DPP thing, though I don't use it). Like little worms, everywhere. Happens to my 5D2 on every image at 1600 ISO and over. I first noticed the effect on a theatre poster that was blown up to 2 metres by about 3 metres. It was just too painful to look at. It's even noticeable on an A2 print.

BUT it doesn't happen with Raw Developer. I love the workflow with LR but just don't rate it's ability to get the best image. Even though RD takes me at least twice as long to use I have now forced myself to (semi) master it and the results are clearly better.

[DavidP]

You are using that test pattern with the lines descending in size, you are almost guaranteed to hit the spot where the sensor sites line up with the target lines and moire happens like crazy.
In real life use you will see in much less, the higher the resolution gets, the less of a problem it is becoming.

[free1000]

Classic aliasing effect, it shows that the 5D2 hasn't got an over strong anti-aliasing filter.

Yes Raw Developer is better at handling this if you get it.

I've seen this also with the 1DsII, but only with a really, really, sharp lens like the 90 TS-E.

I have to say, the resolution on the 5DII is fantastic, the flip side of that detail is that occasionally you have to deal with an artifact like this.

I was noticing on some sRAW2s that I shot that these must be just about the most detailed 5Mp files I've ever seen, certainly more is resolved than I ever saw on the eos 10d, though I have upgraded my lenses since I had that camera.

Another solution at shooting time is to shoot with a very high aperture.  F22 may be enough for a bit of diffraction softening to add a stronger anti-aliasing effect and remove these artifacts, effectively like applying a little softening at shoot time. Hard to do sometimes because you often don't notice the effect at shoot time.

[Panopeeper]

It is a creature of the demosaicing algorythm.

I received the raw files from Richard and took a close look at it.

There is strong moiré in the shots; the strange pattern is the result of the demosaicing of that. The attached captures show the moiré hugely exaggerated.

[madmanchan]

These are classic demosaic artifacts, due to the algorithms' difficulties in finding the "optimal" direction in which to interpolate the missing colors.

[ejmartin]

These are classic demosaic artifacts, due to the algorithms' difficulties in finding the "optimal" direction in which to interpolate the missing colors.

Yup.  Repeated patterns with pixel-level variation are prime causes of these sorts of artifacts. Some algorithms are better than others at coping.  Here's a non-test chart example; ACR 4.6 on the left, experimental demosaic algorithm on the right (warning -- 1.1MB file):

http://theory.uchicago.edu/~ejm/pix/20d/tests/ACRvsAMZ.png

[rwheat]

Thankyou everyone who replied.

Your expertise and the information you provided is very much appreciated.

Richard.

[Panopeeper]

These are classic demosaic artifacts, due to the algorithms' difficulties in finding the "optimal" direction in which to interpolate the missing colors.

Yes, but the underlying reason seems to be the Moiré. Look at for example the second capture I posted, and then Richard's first sample: the criss-cross pattern is exactly there, where the V-shaped Moiré is.