A Biased Evaluation of The Differences...

[Rob C]

Calm, folks, it´s only opinions, cameras and lenses, after all; not as if anyone is forcing anyone else to do anything at all.

John Camp: very well put.

Rob C

[Mort54]

Patronising and ignorant fool.

You really are taking this WAAAY too seriously.

I simply stated that the whole noise difference between Canon and Nikon, tho real, was overated. By that I meant that when you actually printed the images, most of that noise difference wasn't even visible for typical print sizes. What you see at 100% on a monitor isn't the same as what you'll see in a print.

Anyway, enjoy your camera, whatever it is. Whatever deficiencies you feel it might have, or that the cameras of others might have compared to yours, isn't worth loosing sleep over.

Have a good day.

[Quentin]

Patronising and ignorant fool. I don't care what the label on my camera is.  I'll simply use what ever is best for my needs.
Nikon were useless for my needs, now they aren't.  There was no bias in the decision, just a list of abilities the Nikon did not have compared to the Canon.
Fanboys are so pathetic.
[a href=\"index.php?act=findpost&pid=168811\"][{POST_SNAPBACK}][/a]

I think that maybe by using language straight out of dpreview, your reputation might just have gone futt futt futt.

[Mort54]

I think that maybe by using language straight out of dpreview, your reputation might just have gone futt futt futt.

Well, in fairness to jjj, I was being a bit of a smart aleck earlier in the thread, so I'll accept my share of responsibility for some of the tone this thread has taken on (of course, a lot of others have contributed to this tone, so I don't accept full responsibility :-)

[Ray]

Setting aside the issues which we can only speculate upon, such as how much of the noise will be visible in a print of a certain size; how much the results would change using a different RAW converter; what the consequences would be of different settings in different noise reduction programs etc.... Does anyone else see what I'm seeing in the Ken Rockwell images?

Namely:

(1) The 5D images (at ISO 25,600) have much more obvious chroma noise which can, however, be easily removed in Noise Ninja with no further loss of resolution.

(2) The 5D image is stuck with a much coarser type of luminance noise which is difficult to remove without destroying detail.

(3) Despite the 5D's more obvious luminance noise, the curly structure and texture of the doll's fabric is more apparent than in the D3 shot (at ISO 25,600).

(4) The D3 image looks cleaner and has an appearance of greater sharpness, but it's a false sharpness. The image has been messed up to a degree with in-camera processing.

(5) Attempting to further improve the D3 image through noise reduction programs results in obvious softening. The noise reduction has already been done in-camera.

(6) The 5D image can be improved to a greater extent using noise reduction programs, particularly with regard to chroma noise.

Taking all these factor into consideration, the actual noise advantage of the D3 for RAW shooters (compared with the 5D) seems fairly marginal to me and claims of a 2 stop advantage seem fanciful.

If Canon can come up with an upgrade to the 5D which produces the same image quality at ISO 6400 as the current model produces at ISO 1600, then they don't even have to increase the pixel count for me. I'll buy it   .

[jeffok]

Setting aside the issues which we can only speculate upon, such as how much of the noise will be visible in a print of a certain size; how much the results would change using a different RAW converter; what the consequences would be of different settings in different noise reduction programs etc.... Does anyone else see what I'm seeing in the Ken Rockwell images?

Namely:

(1) The 5D images (at ISO 25,600) have much more obvious chroma noise which can, however, be easily removed in Noise Ninja with no further loss of resolution.

(2) The 5D image is stuck with a much coarser type of luminance noise which is difficult to remove without destroying detail.

(3) Despite the 5D's more obvious luminance noise, the curly structure and texture of the doll's fabric is more apparent than in the D3 shot (at ISO 25,600).

(4) The D3 image looks cleaner and has an appearance of greater sharpness, but it's a false sharpness. The image has been messed up to a degree with in-camera processing.

(5) Attempting to further improve the D3 image through noise reduction programs results in obvious softening. The noise reduction has already been done in-camera.

(6) The 5D image can be improved to a greater extent using noise reduction programs, particularly with regard to chroma noise.

Taking all these factor into consideration, the actual noise advantage of the D3 for RAW shooters (compared with the 5D) seems fairly marginal to me and claims of a 2 stop advantage seem fanciful.

If Canon can come up with an upgrade to the 5D which produces the same image quality at ISO 6400 as the current model produces at ISO 1600, then they don't even have to increase the pixel count for me. I'll buy it   .
[{POST_SNAPBACK}][/a]

Have you seen this fellow's review of the 1Ds III? [a href=\"http://www.juzaphoto.com/eng/articles/canon_eos_1ds_mark3_review.htm]http://www.juzaphoto.com/eng/articles/cano...ark3_review.htm[/url]
I think he makes a similar conclusion  as you do about the real difference in noise compared to the D3 and Canon 40D. Seems reasonable to me.

[Ray]

Have you seen this fellow's review of the 1Ds III? http://www.juzaphoto.com/eng/articles/cano...ark3_review.htm
I think he makes a similar conclusion  as you do about the real difference in noise compared to the D3 and Canon 40D. Seems reasonable to me.
[a href=\"index.php?act=findpost&pid=168929\"][{POST_SNAPBACK}][/a]

Wow! I think John Sheehy might have some comments to make on that review. The 1Ds3 and 5D are about equal regarding noise at ISO 3200, even when the 1Ds3 file has been downsampled to the same size as the 5D image?

[Craig Arnold]

It does seem that the D3 is doing something slightly different (from what they or anyone else) has done before in terms of noise reduction as the data is read off the sensor and before the RAW image is written.

I'd be interested to know from the experts how much software intervention the other cameras do and what Nikon might be doing differently.

I agree with the observation (from other peoples tests - I don't have a D3) that the Canon images clean up nicely with some noise reduction and the D3 doesn't, and that after that NR on the Canon images the difference seems less than 2 stops. But really even a 1 stop improvement on the 5D is pretty impressive.

I also note with interest that the new version of DXO v5 (which I haven't tried yet because the Mac version is still under development) claims to do some interesting things in NR before demosaicing in the RAW conversion process. DXO make a lot of their money from selling software for use inside the digital processing chips of various manufacturers; is it possible that Nikon are doing something similar or have simply bought the DXO technology for use in the D3?

I would suggest that the obvious downside to the in-camera NR is that if it does reduce detail when it reduces noise that of course the user doesn't get to choose what he prefers. Less work in post, but ya gets what ya gets.

[John Sheehy]

Wow! I think John Sheehy might have some comments to make on that review. The 1Ds3 and 5D are about equal regarding noise at ISO 3200, even when the 1Ds3 file has been downsampled to the same size as the 5D image?
[a href=\"index.php?act=findpost&pid=168933\"][{POST_SNAPBACK}][/a]

I haven't looked at the article yet, but I do have this:



This is the 1Dsmk3 at ISO 51000, downsampled to 77%, 100% crop, manual conversion from RAW, with a little sharpening.  No NR, whatsoever.  To simulate NR, squint when you look at it.

How does this compare to a 100% crop from the 5D or D3 at ISO 51,000?

[Graeme Nattress]

There is a distinct advantage to apply chroma noise reduction on RAW un-demosaiced data, compared to applying it to post-demosaic RGB data. If that's the difference we're seeing here, I'd not be surprised one little bit. Once you demosaic, and especially after colour space conversion, you co-mingle the RGB data in such a way that interferes with noise reduction targeted at the exact bits of raw data that are causing issues.

Graeme

[sojournerphoto]

There is a distinct advantage to apply chroma noise reduction on RAW un-demosaiced data, compared to applying it to post-demosaic RGB data. If that's the difference we're seeing here, I'd not be surprised one little bit. Once you demosaic, and especially after colour space conversion, you co-mingle the RGB data in such a way that interferes with noise reduction targeted at the exact bits of raw data that are causing issues.

Graeme
[a href=\"index.php?act=findpost&pid=168978\"][{POST_SNAPBACK}][/a]

Graham, what sort of algorithm is used to apply chroma nr to raw data? is this another smoothing/low p[ass type function or something else all together?


just interested

Cheers

Mike

[Graeme Nattress]

I'm not at liberty to divulge that kind  of information. Sorry. I can just tell you that it works very well indeed, but I can't get into specifics of  exactly how it works. I do a lot of R&D in this area.

Graeme

[sojournerphoto]

I'm not at liberty to divulge that kind  of information. Sorry. I can just tell you that it works very well indeed, but I can't get into specifics of  exactly how it works. I do a lot of R&D in this area.

Graeme
[a href=\"index.php?act=findpost&pid=168986\"][{POST_SNAPBACK}][/a]

That sounds liek the old 'if I told you I'd have to kill you!', but actually I believe you. I must admit that I do find this type of image processing stuff interesting.

Mike

[Ray]

How does this compare to a 100% crop from the 5D or D3 at ISO 51,000?
[a href=\"index.php?act=findpost&pid=168974\"][{POST_SNAPBACK}][/a]

How can we say, John? We haven't got the 5D image to compare   .

[jjj]

I think that maybe by using language straight out of dpreview, your reputation might just have gone futt futt futt.
[a href=\"index.php?act=findpost&pid=168863\"][{POST_SNAPBACK}][/a]

I have no longer have much patience for people who cannot read posts and post a lot of nonsense whilst also misrepresenting what I say. So if someone doesn't want to get called an idiot or whatever, then they shouldn't behave like one. Though to give kudos to mort54, he has admitted to being a bit of a smart alec and provoking the response, so at least I respect him for saying that.

[Ray]

I agree with the observation (from other peoples tests - I don't have a D3) that the Canon images clean up nicely with some noise reduction and the D3 doesn't, and that after that NR on the Canon images the difference seems less than 2 stops. But really even a 1 stop improvement on the 5D is pretty impressive.
[a href=\"index.php?act=findpost&pid=168942\"][{POST_SNAPBACK}][/a]

Peripatetic,
My test images indicate the D3 is definitely not as much as a 1 stop improvement on the 5D in terms of noise at high ISO. What I think is happening here is that many photographers, both amateurs and professionals, are getting bamboozled by the unprecedented high ISO numbers.

An exposure at ISO 6400 is not necessarily one stop less than an exposure at ISO 3200. The ISO numbers are only an indirect and approximate indication of what the actual exposure might be.

For example, a D3 shot at F11 and 1/30th is one stop less than a 5D shot at F8 and 1/30th. I can show you two such shots and several 100% crops. I think you'll agree the D3 shot is clearly noisier.

I mentioned earlier that i thought the D3 noise advantage over the 5D is of the order of 1/4 to 2/3rds of a stop. Having now examined more images from both cameras, I believe it is actually less. More like 1/4 to 1/3 of a stop.

The final image compares the 5D at F8 and 1/50th with the D3 at F11 and 1/30. That's 1/3rd of a stop difference. I still think the 5D has a slight edge here, but there's no point in arguing whether it's really 1/4 stop or 1/3rd. Let's just say it's definitely less than one stop and apparently less than 1/2 a stop.

[attachment=4820:attachment]  [attachment=4821:attachment]  [attachment=4822:attachment]

[attachment=4823:attachment]  [attachment=4824:attachment]  [attachment=4826:attachment]

I should mention that none of the above images have been resampled. The 5D image is slightly larger because the file size is larger (72.8MB as opposed to 69MB for the D3). There is also a slight discrepancy in focal length due to inaccuracies of the zoom labelling, but only slight. This should not affect noise results to any noticeable extent.

[John Sheehy]

How can we say, John? We haven't got the 5D image to compare   .
[a href=\"index.php?act=findpost&pid=168997\"][{POST_SNAPBACK}][/a]

It would be nice if someone shot the exact same scene with the same lens and Av and TV values with all three cameras, but no one who isa ble to do so seems interested in doing so, and providing the RAWs.

In general, though, how does this compare to a 100% crop at ISO 51000 on the other cameras?

This image reminds me; when I do these manual conversions with no fancy demosaicing, the texture always looks much better to me than the mess you see in most converters at 100%; they seem bent on bringing out high-contrast detail at the pixel level, and for the most part, it just looks like artifacts; things like pieces of broken straw baskets with the straw painted randomly.  All the more reason, IMO, to move to higher pixel densities, and oversample the lenses in the red and blue CFA channels.  Say goodbye to demosaicing - a necessary evil of low-res cameras.

[Mort54]

I have no longer have much patience for people who cannot read posts and post a lot of nonsense whilst also misrepresenting what I say.

Not to reopen old wounds, but I hope you aren't suggesting I misrepresented what you said, because I never did. I did make some comments that were intended as good natured needling (my "smart aleck" remarks), but I never misrepresented you. Also, I have to say that even tho you may have felt provoked by some of my comments, you strike me as someone who is looking to be provoked, or at least someone with a chip on his shoulder. Several of your posts seemed very combative right from the get-go, and you made several needling remarks of your own, which is why I responded to you the way I did. I really have nothing further to say on the matter. Hopefully this forum won't degenerate into the rude free-for-all we see on DPReview and other sites.

[John Sheehy]

Graham, what sort of algorithm is used to apply chroma nr to raw data? is this another smoothing/low p[ass type function or something else all together?
just interested
[a href=\"index.php?act=findpost&pid=168982\"][{POST_SNAPBACK}][/a]

One simple method is a CFA-aware filter.  The filter knows that the RAW data is actually 3 or 4 interleaved, color-filtered greyscale images, while it is still in a mosaic form, and performs a median filter or high-pass subtraction or whatever on the color channels by themselves.  Then, lacking wild excursions of the original noise, the subject has a more consistent color from pixel to pixel.  A more advanced algorithm might detect edges, and avoid softening them.

[sojournerphoto]

One simple method is a CFA-aware filter.  The filter knows that the RAW data is actually 3 or 4 interleaved, color-filtered greyscale images, while it is still in a mosaic form, and performs a median filter or high-pass subtraction or whatever on the color channels by themselves.  Then, lacking wild excursions of the original noise, the subject has a more consistent color from pixel to pixel.  A more advanced algorithm might detect edges, and avoid softening them.
[a href=\"index.php?act=findpost&pid=169090\"][{POST_SNAPBACK}][/a]

Thanks John

based on an immediate response, it seems that such an approach has the potential in some circumstances to remove signal as well as noise? Presumably the more advanced algorithms would be able to better differentiate, but anything close to the sampling limit will be vulnerable?

On another note, responding to your post on much higher resolution sensors, I tried and rejected photo acute as I found the 'additional' detail just looked like artifacts to me. I suspect the answer is more resolution that would allow a proper sampling of all the lens can project - diffraction abberations and all. Plus with enough resolution the spcial resampling necessary to address lens defects a la DXo would be less destructive.

Mike