p.3 #1 · Ming Thein on pixel density and sharpness
Thanks Suede for clearing this all up.
Makes sense.
This article by Ming is a continuation from his D800E midterm review where he lists all the issues he sees with shooting the D800E.
One of those is the higher shutter speed 1/2x focal length to get non blurry results at the 100% pixel view when shooting handheld.
He states he has not bonded with the D800E like earlier cameras like D700 and prefers to only use it for commercial work and not for his personal shooting.
He also mentioned that he prefers the D800E over the D800 because it is perceptually sharper at narrow apertures like f11 which are being affected by diffraction. Ming shoots a lot of macro of watches.
I love my D800E and agree with the 1/2x focal length rule for when I am handholding, but usually don't mind just bumping up the ISO just a stop to achieve that extra shutter speed. The high ISO noise performance is very good.
Ming also mentions the AF not being as solid and the lack of focus screens as other reasons.
Some people like more focus snap, but I can focus my MF Zeiss and Leica glass fine on my D800E just like my 1ds3 with stock screen.
p.3 #2 · Ming Thein on pixel density and sharpness
It seems a bit arbitrary to say, you're scaling down images from X to Y for no reason whatsoever.
More likely where it matters, you're scaling down images to whatever web size you want to post on (which hopefully won't be modified by whatever hosting service you use).
Or where it's critically important, how does X megapixels translate into a given print size.
p.3 #3 · Ming Thein on pixel density and sharpness
carstenw wrote:
So let's take the D800, which has an AA filter. Are you saying that if you downsize a 36MP photo to a 35MP image, it will be sharper than a native 35MP camera would be? Sharper in terms of resolution or perception? I think Ming Thein is only talking about perception.
All else equal, the downsized image has the potential of being of higher fidelity.
p.3 #5 · Ming Thein on pixel density and sharpness
FlyPenFly wrote:
It seems a bit arbitrary to say, you're scaling down images from X to Y for no reason whatsoever.
More likely where it matters, you're scaling down images to whatever web size you want to post on (which hopefully won't be modified by whatever hosting service you use).
Or where it's critically important, how does X megapixels translate into a given print size.
Yes, all these things including the large file sizes are too big for efficient work flow become his reasons/excuses not to use his D800E for personal shooting instead of his OM-D.
It is easy to see from his blog that he shoots a lot of street, and of course smaller, less intrusive cameras like OM-D or Leica M are better suited. I think he needs the new Leica which would give him excellent IQ and all the other benefits.I bet he will upgrade from his 9 soon.
p.3 #6 · Ming Thein on pixel density and sharpness
Toothwalker wrote:
All else equal, the downsized image has the potential of being of higher fidelity.
Sure, there is theoretically more information there, but I am not talking about potential, but reality. We have to deal with imperfect downsizing algorithms, and my feeling is that when you downsize images with real software, for a while you lose sharpness, but after a while, you gain it back.
Another way to think about it. Slightly rotate an image. In theory, there is the same amount of information there (take or give a tiny amount), but in reality, as soon as you rotate a little, some softness creeps in.
p.3 #7 · Ming Thein on pixel density and sharpness
TheSuede: Thank you for your excellent post.
Has it been determined, how big the amount of false image information due to aliasing on a AA-less camera is? Say for a combination of the D800E with an excellent Zeiss lens? It's clearly dependable on the scene and many other factors, but the information that is resolved by the lens but not by the sensor, should add some kind of noise to the pic.
Carsten: With this small amount of downsizing, the errors due to the algorithm should prevail the benefit of averaging.
p.3 #9 · Ming Thein on pixel density and sharpness
carstenw wrote:
Another way to think about it. Slightly rotate an image. In theory, there is the same amount of information there (take or give a tiny amount), but in reality, as soon as you rotate a little, some softness creeps in.
p.3 #10 · Ming Thein on pixel density and sharpness
carstenw wrote:
Sure, there is theoretically more information there, but I am not talking about potential, but reality. We have to deal with imperfect downsizing algorithms, and my feeling is that when you downsize images with real software, for a while you lose sharpness, but after a while, you gain it back.
Another way to think about it. Slightly rotate an image. In theory, there is the same amount of information there (take or give a tiny amount), but in reality, as soon as you rotate a little, some softness creeps in.
You can deal with imperfect downsizing algorithms if you want, but I'll take my chances.
p.3 #13 · Ming Thein on pixel density and sharpness
There is no perfect resizing algorithm, due to the square base shape of a "pixel". Aliasing vs smoothness has to be balanced by some kind of average image estimation.
You can do a very good information-preserving downsample in PS, I use this action
1) convert image to first 16-bit, then linear gamma ARGB (there are settings for thins in "custom RGB" in the "Edit>Convert to profile" in PS)
2a) If you're going to downsample by just a bit, first enlarge with standard PS biCubic to 2x intended size
2b) If you're downscaling to <50% but >25%, first downsample with standard PS biCubic to 2x intended size
3) Run either a blur on the image, either:
3a) 0.5px gaussian for less aliasing, or
3b) filter>other>custom 2x2 box blur for a sharper image
4) scale to 50% with nearest neighbor chosen as algorithm.
I've set F12 as "halve by square blur" and shift-F12 as "halve by Gauss blur".
F11 is normal Adobe RGB, shift-F11 is Adobe RGB 16 bit gamma 1.0
If you're scaling to 25% or less, don't use square filter on all halvings, then the result will be fully "perfect square pixel presentation". This isn't visually pleasing though it is the perfect mathematical presentation.
If you do this, then a 95% scaling will be very close to perfect, without any softening effect.
You can do this with "just" long-pole-sinc or Laczos also, and get a good data fidelity - but it will be more aliased. And you can't do it in PS.
