I asked Thom Hogan on this.
His response:
Answers below
--
Thom Hogan, writer/photographer
bythom.com, dslrbodies.com, sansmirror.com, zsystemuser.com
- Is it in reality a better idea to cleanup noise in a raw vs a photoshop/Tiff?
Yes, if the algorithm is good. The problem is that when you demosaic an image, you use near-neighbor values to create a specific RGB pixel. That means that noise is encoded from the neighbors as well as the original pixel. This can be subtle, but it’s a real issue.
> DxO claims they know each lens the weaknesses and apply different sharpening across the frame Vs. ordinary unmasked sharpening. To me this is just fancy marketing. What’s your opinion on this?
No, it’s not fancy marketing. The whole thing about DxO from the beginning has been that they actually take product and measure it. Again, it’s subtle, and your sample of a lens may vary some from theirs, but lenses (and thus images) are not invariable in blur circle from edge to edge.
> It is also confusing their claims, Soft, sharpening = 0 Standard = 1 = 2 = 3 Using soft, isn’t applying their magic lens soft treatment?!
That you’d have to ask them for details about. This is the usual “the engineers now what they did, but the marketing team either is trying to simplify that to absurdity or didn’t understand it themselves.
> So, Only DxO measures the amount to correct, while Adobe uses gauss work?! Give me a break…
No, DxO actually does their own measurements. Adobe uses (for the most part) the camera companies “profiles.” It’s those profiles that are problematic. For some lenses they’re overly simplistic, which is one reason why you sometimes see residual linearity issues even after Adobe (or the camera maker) applies a profile. This specifically shows up with lenses with mustache distortion, because the profiles only define simple barrel or pincushion adjustments and where to make those.
> Other then the “Speed” of their newest deep prime xd2, I couldn’t find any difference between DxO and the latest ACR Denoise AI.
Yeah, well I don’t know what you can and can’t see. I don’t know if you’re trained to see what you’d need to be trained to see. I discount pretty much all claims of “I don’t see a difference.” Some of us measure, some of us have been trained (heck some of us are trainers). No doubt that we’re up to the point where we’re chasing small differences, and most people can’t see small differences, and most uses of an image wouldn’t come close to reveal those small differences.
> I tried pixel peeping in LRC on a number of different pictures, under different conditions, different subjects. From portraiture to wildlife. I couldn’t detect any differences.
Note what I just said, but also know this: LRC (and most products) don’t ever really show you actual pixels. They scale to the display, and many, including Adobe’s use math I’d call “challenged.” That’s because they want to do what they do fast, not accurately (which takes time).
> The only real difference I can see, and I like it, is NX Studio… I love the randor of the colors. I could never get Adobe to really match it.
Note what I said about trained versus trainers. I can match Adobe to NX Studio. It’s not simple and it’s not something that can be applied across all images (e.g. a preset), but it can be done. But also, we’re back to that math thing and the display. If you take shortcuts with the math, which Adobe does, you get different results than if you use floating point correctly.
> And the free Noise Reduction in NX, is VERY effective and the most natural of them all.
Not in my judgement. It’s okay, but it really has an interaction with sharpening that can make it stand out.
> It still looks good and pleasing to eye.
Pleasing to YOUR eye. That’s actually one of the issues facing everyone. We have varying degrees of what we can see and what we like. By all means, choose what you like, but don’t for a second believe that’s a “standard” and that everyone will like it.
> Sadly NX is behaving like an outdated software on an outdated computer….
Well, it is. NX Studio is actually Silkypix—and I believe an older version—on which Nikon has mounted their JPEG rendering engine.
