Oh dear....5D3 vs D800

<blockquote>

<p>In any case it's pretty difficult to apply any sensible NR to bayer sensor raw data since it would first have to be converted to RGB then noise reduction applied, then back to the original raw format - doesn't make much sense to do that.</p>

</blockquote>

<p>That's a very interesting point, Ilkka.</p>

<p>Dan, no, I find noise reduction pretty useless when it comes to printing. I tried it, & it just blurred detail (especially in shadows, where I was concerned about noise) too much... make it soft/muddy. Even in my film scans, while noise reduction makes the image look great on screen, it rarely matters for print. If anything, the grain ends up making prints look sharper. Therefore, I tend to not indulge in NR much.</p>

<p>I recently did a bunch of Stouffer Transmission Wedge tests to estimate DR based on max signal vs. the patch on the wedge where SNR drops to 1. I get 11.3EV for the 5D Mark III & >13EV for the Nikon D7000.</p>

<p>But this test is a little hard for me to interpret since the mean of the read noise on the D7000 images is ~0.43 ADU... so even for the darkest patch on my wedge (which has 13.2EV DR), when the brightest patch still hasn't blown out the sensor, the SNR for the darkest patch is well above 1 for the D7000 (signal = 3.11ADU). But what I'm curious about is: is the Nikon image still maintaining the relation between dark patches? I believe I need to fit the signal of the patches to a logarithmic curve & then assess the deviation... then do the same for the Canon... to see if the cameras are *accurately* representing the densities of the patches.</p>

<p>Anyone have experience with this? Maybe I should just purchase Imatest & run it through there rather than do all these quantitations manually? Still, I feel more comfortable when I know exactly the math behind everything... :)</p>

<p>Thanks,<br />Rishi </p>

<blockquote>

<p>If someone do need to print A1 or A0 sizes then the D800 is a must-have.</p>

</blockquote>

<p>Of course. Everyone who wanted to print at these sizes prior to the emergence of this greatest thing since sliced bread had to go MF.</p>

<p>Happy shooting,<br>

Yakim. </p>

<p>If you do noise reduction for large prints, here are some thoughts about how to approach it:</p>

<ul>

<li>Be selective about where you apply it. Using masked layers or other techniques, keep it away from areas of great detail (where the noise won't be noticed as much anyway) and apply it sparingly to areas of less detail that are of uniform luminosity - the "loss of detail" there won't matter.</li>

<li>Don't even think about trying to "eliminate" noise. It isn't possible and most printers will agree that it isn't even desirable. If you can produce a completely noise free image you are also likely to produce a plastic-looking image. Some small amount of noise (of appropriate quality) tends to produce a more appealing result than one with "no noise."</li>

<li>Surprisingly, one way to deal with certain types and levels of noise is to... add more noise! By adding a neutral gray "overlay" layer and applying a very small amount of noise you can create a bit of a (he shudders as he types it!) "film like" look that is quite appealing and which will mask some other sorts of noise.</li>

<li>Don't worry about noise that you can find if you go looking for it at 100% or larger magnification. <em>Make a print</em> or a portion of a print at your target size and look at it to see if the noise issue is really important or not. It probably won't be.</li>

</ul>

<p>I'll tell a story on myself in this regard. I shoot with a 5D2. As some know, there is a story about this camera producing "noise banding" in shadow areas, especially if you radically fill or otherwise lift very dark tones in areas of uniform luminosity or gradual gradients. Let me tell you that you can very rarely find some small degree of "banding" in the noise in these areas if you go looking for it at 100% under the right conditions. (I can locate it - if I look for it - it far less than 1% of my photographs - but it is emphatically not a problem in actual photographs that I print.)</p>

<p>A couple years ago I made a photograph along the coast north of San Francisco that included a deeply shadowed foreground cliff face and, beyond that, the sunlight reflected directly off the surface of the ocean. This is just about the worst case scenario in terms of dynamic range - exposing for the shadowed cliff face would badly blow out the reflections, and exposing to control the reflections would result in a very underexposed cliff face. Making this more complex, the way that the moving water was included in the composition made exposure blending (which I use a lot) just about impossible. </p>

<p>I had one frame that I liked a lot. The highlights were controllable with a bit of "recovery" in Photoshop, but the cliff face was so dark that in the large print I was creating I was in danger of ending up with a very large area of blocked shadows - not a good thing! I needed to lift the darkest tones on that cliff face enough to restore the sense of some detail there. So I went to work with black point, fill, and curves to get things where I wanted in the cliff shadow - and when I had the right tones, I could see some degree of banding in the shadow noise, along with a bit of "regular" banding from lifting the very dark tones. </p>

<p>No problem. I duplicated the image layer, created a group of layers, added an overlay, and added a small amount of noise to that overlay. Then I set the mask for this group to be "hide all" and then "painted in" bit of the noise in the deep shadow areas. The banding (both types) was muted by this additional bit of noise. I made a print and it looked just fine.</p>

<p>But the story isn't over. For fun, I went back and cancelled this "noise layer" that I had used successfully to mute the noise that I was worried about and then made another print without all of this compensation - just to see what the difference would be. While I could, indeed, now "see" the noise problem more clearly at 100% on the screen... <em>it was completely invisible in the actual print!</em> Needless to say, after all that work dealing with a "noise problem" that turned out to be essentially theoretical, I ended up using the version in which I turned off the layer than was intended to mask the noise.</p>

<p>Dan</p>

Rishi,

I am trying to understand your post which is quite technical and challenging given my port knowledge in these matters.

Would you be able to explain what "SNR drops to one" means ? Also what does ADU stand for ?

Should I understand that what you are trying to understand is not just at which point noise covers useful signal, but also

how a sensor differentiates between different shades of shadow above the "noise level" ?

If you can direct me to easily understandable reading (for the uninitiated) on these issues, I would be very thankful.

<p>Paulie Smith wrote something about:</p>

<blockquote>

<p><em>"... ultimate image quality..."</em></p>

</blockquote>

<p>"Utimate" image quality? Really?<br>

Dan<br>

-------------------<br>

Dan, shoot a dozen or so images with each camera side by side in the same conditions with similar ISO and whatnot. Then make 16x20 or so prints of each. Set them up side by side and let any number of people view them to compare and see if either body consistently gets picked out as 'finest quality'. That is what I mean by "ultinate image quality' - the best you can get from each. Not a mythical standard but the finest print where you can see, not imagine, the difference.<br>

My standard is 8x10 contact prints and I am admittedly a stick in the mud in many ways. If I can't see the result in the final image I don't change things. That difference does include how one makes the image - if the equipment itself makes it more likely to get the result compared to another body. We know both Canon and Nikon make some very good gear. Nikon metering and Nikon strobe control beats canon like a red headed stepchild. The only real question for me these days is whether the images the body can produce are images I can live with. Given the choice I would pick the body that gave me higher quality - however you define it - and images where I can actually see that higher quality.<br>

I understand well that 'ultimate image quality' is at times not needed nor is it wanted. You shoot and print what is wanted, not a technical chart of comparison lines. You choose lenses and settings specifically for the 'feel' rather than ultimate sharpness or whatever. That is interpretation and artistic license. But failing to use the best gear, technique or materials you can and trying to excuse it for 'art' is stupid. Most of our cameras and lenses are capable of a lot better work than we often produce. I would choose the body that gave me the best chance of creating my vision any day - within the limits of bank account and real world usability.<br>

The sensor in one body is superior while the Af and frame rate in the other is superior. Horses for courses and we are lucky to have a choice -and Sony and others pushing more and better development every day.</p>

<p>Paulie:</p>

<p>I print in-house to 24" wide on an Epson 7900. Shooting a 22MP 5D2 and then making large prints, there are several things I know about how this all plays out.</p>

<ul>

<li>There is no "ultimate" anything in photography, especially in the technical realm. There are various specification levels that may make one current camera better or worse than some other current camera for your particular purpose, but none of them approach whatever "ultimate" quality might be. Camera technology already has (arguable) about 140 years of continuous improvement, and this will continue.</li>

<li>When used by with care and skill by competent photographers, current full frame DSLRs can easily produce stunning prints at 18" x 24" size and very, very good ones at 24" x 36". If one does not regularly print this large, among the factors that might make a camera more or less right for you, photo site density is way down the list if it is on it at all.</li>

<li>Once we push full frame DSLR photograph to these sizes and beyond, there are a range of factors that begin to have critical effects on the quality of the print. While photo site density can be among them, it is often not the limiting factor. At least as often the limit may be things like absolute critical focus, camera stability, DOF issues, and so forth.</li>

<li>Even when all of those things are well-handled, the actual difference in image size at which the same ppi resolution can be maintained is not as much larger with 36MP compared to 22MP as many might assume. It is a matter of inches - and nothing at all like, say, doubling print size.</li>

<li>That said, if I were not committed to another brand right now and I were getting my first FF body and I was a very skillful and careful shooter and I was going to regularly print high quality images at 24" x 36" and beyond, I would certainly consider the excellent Nikon D800.</li>

<li>On the other hand, I cannot see the real world value of a brand switch to get this larger number of photo sites for few months before Canon comes out with something that will compete with it, as Canon surely will (have to) do.</li>

</ul>

<p>And, as you mention at the end of your note, all of these cameras are very, very fine photograph making machines. </p>

<p>Your comment that "falling to use the best gear, technique or materials you can and trying to excuse it for 'art' is stupid" perplexes me. By "you" I hope you are not referring to me, although you did address your comment to me. Higher quality is defined in a number of ways. To my way of thinking, a marginal or invisible "difference" in one of several measurable technical qualities is most certainly not compelling as a reason to move from one brand to another. Keep in mind, as well, that lenses have something to do with "quality," as does the photographer's learned skills with the particular gear when it comes time to making the shot.</p>

<p>Better is better if it makes my work better. Better is merely interesting if it has no significant positive effect on my photography. </p>

<p>Dan</p>