A simple demo of the benefits of resolution…

[ErikKaffehr]

Hi,

Please note: The sole purpose of this posting is to demonstrate the benefits of high resolution over say the 39 MP the P45+ delivers. The stitched Sony image is about the same size as P45+ and resolution is similar to the IQ3-100MP. So it actually tells exactly what you would see with a Planar 100/3.5 on a IQ3-100MP.

There used to be a lot of discussion about how much resolution is needed. My personal opinion is that the resolution needs to match the resolution of the lens.

I have two cameras with different resolution, a Hasselblad P45+ and Sony A7rII. In this case I shot with the same lens, a Planar 100/3.5 that is known to be one of the sharpest lenses on both cameras. I used my HCam Master TSII with the Sony, allowing something like +/- 10 mm shift, resulting in a 44x36 mm image (somewhat larger than the IQ-250 and somewhat smaller than the P45+). The limitation here is the internal shading from the HCam Master TS.

So we can compare the two images side by side, the first crop shows a central crop, both images at actual pixels. Check the road sign on the P45+, it has a lot of colour aliasing, while the stitched image from the A7rII is clean. The planar lens outresolves the Sony sensor, the reason that it has no visible aliasing are probably the gapless microlenses on the Sony A7rII. Still the Sony can alias, too. Also check on the blinders above, P45+ has a lot of color moiré while the Sony A7rII resolves the actual detail.

The next image I have scaled down the stitched Sony image to the same hight as the P45+ image, just using image size in Photoshop CC.

Very clearly, the downsized image is still much clearer than P45+ image.

Finally I took another crop from the downsized image, using some tree tops. In this case the P45+ image is quite diffuse while the stitched Sony A7rII image is contrasty and sharp. The explanation here is probably that interpolation in demosaic destroys the fine detail on the P45+, while the smaller pixels on the Sony A7rII still resolve enough detail.

So, what is my take? The stitched image from the Sony sensor is clearly preferable to the P45+ image. So a 1.1X crop sensor can do well with 77 MP and 100 MP (or so) would be just fine on a full frame 645 sensor.

Another observation is that even when scaled down to 39 MP (or so) the 77 MP image from the Sony holds it's own over the P45+ image. Better to sample properly and down-sample than under-sample to begin with.

Some good lenses from the Victor Hasselblad era will held up well to 100 MP sensors on full frame, some obviously won't.

Best regards
Erik

[Christoph B.]

Now compare the Sony to a stiched P45+ photo!

[ErikKaffehr]

Hi,

The aliasing stuff would not be solved by stitching the P45+. The only solution is a better sensor....

Or, it would have been possible to use a longer lens (like Sonnar 150/4) stop it down to f/16 to reduce resolution and stitch.

But, this test was not about bragging rights, you know. It was about the benefits of high resolution sensors on MF, and that can be demonstrated using a lens I bought used for 700$US, a 3200$US camera body and a 1200$US TS adapter, instead of buying a 40000+ $US back from Phase One.

What the images here show, is what you can gain from a 100 MP back compared to a P45+.

Best regards
Erik

Now compare the Sony to a stiched P45+ photo!

[Christoph B.]

Hmm I actually think it would be!

Use a longer focal length, you'll get a bigger magnification and more detail and the aliasing won't be an issue any more.

I know what you mean but stitching isn't always the way to go - after all you can also stitch P45+ files - and by doing so (with an equivalent of a 100mm on small format) you'll definitely get better results. I know that the newer sensor has a higher pixel density and will produce a higher resolution for the same surface and stitching will increase the overall resolution as well - but if you can use the one camera for stitching you can also use the other. And downsizing a higher resolution image will always beat a lower resolution image (using the same lens etc).

I agree a higher resolution for a 1.1 sensor would be definitely possible and that a lot of Hasselblad lenses will probably be up to the task of 100mpx - though in this case you're getting mainly the sweet-spot of the center of the image, not the corners.

Remember: the 40.000$ back will deliver the resolution in a single shot and does not require stitching

[ErikKaffehr]

Hi C+B,

Aliasing is really the result of mismatch between the resolution of the lens and the sensor. So, if you have a high resolving lens you also need a high resolving sensor to do it justice. Otherwise aliasing will result.

Now, aliasing may not be obvious. It will always create fake detail as soon as the subject has high frequency detail, and that fake detail may actually enhance the image, as it shows detail that the sensor cannot resolve.

I don't see the need for 100 MP in my images, I don't print that large. But, I felt that I had a good way to demonstrate that small pixels and high resolution do have benefits on MFD, without having the gear. What I say is that resolution is a good thing even if the end result is scaled down.

The image below is a good example. The scale at the bottom has a sine pattern with increasing frequency. Resolution limit is around 74 lp/mm. I think I was a bit closer than the needed 5 m, so check around the 90 mark instead. You see that the lens cannot resolve more, so it becomes essentially grey. Now, moving to the right you see a line pattern again. All of that is fake detail. The sensor cannot resolve all the detail from the lens, so it converts the real detail from the lens into low resolution artefacts.

Best regards
Erik

Hmm I actually think it would be!

Use a longer focal length, you'll get a bigger magnification and more detail and the aliasing won't be an issue any more.

I know what you mean but stitching isn't always the way to go - after all you can also stitch P45+ files - and by doing so (with an equivalent of a 100mm on small format) you'll definitely get better results. I know that the newer sensor has a higher pixel density and will produce a higher resolution for the same surface and stitching will increase the overall resolution as well - but if you can use the one camera for stitching you can also use the other. And downsizing a higher resolution image will always beat a lower resolution image (using the same lens etc).

I agree a higher resolution for a 1.1 sensor would be definitely possible and that a lot of Hasselblad lenses will probably be up to the task of 100mpx - though in this case you're getting mainly the sweet-spot of the center of the image, not the corners.

Remember: the 40.000$ back will deliver the resolution in a single shot and does not require stitching

[Christoph B.]

Wait - you didn't use CaptureOne?

I'm pretty sure the results regarding aliasing will be much better when you work with CaptureOne instead of using Lightroom or Photoshop CC as a RAW converter.

And I fully agree on your point that more resolution is good. Well - if you actually need it and have enough processing power to work with the files. If you don't need it the downsizes files will surely look better, no discussion here.
The only 'problem' I can see is that there's no guarantee the lens will deliver more resolution, I think a lot of lenses from the Hasselblad lineup will work but probably not all of them. As I said, right now you're getting the sweetspot of the center image - but on a 100mpx fullframe it might not look as good in the corners...

[ErikKaffehr]

Hi,

Capture One is better at suppressing aliasing artefacts, but not in this case, see attached conversion.

The way I test, with shifts, I cover a large part of the image area. But, you are right, only a few of Hasselblad's lenses cover the full area with high MTF. The two Sonnars and the Planar 100/3.5 that I have owned. Distagons don't do that.

But, Phase One and Hasselblad claim their new lenses are great, so I am pretty sure they also work well with high res sensors.

Best regards
Erik

Wait - you didn't use CaptureOne?

I'm pretty sure the results regarding aliasing will be much better when you work with CaptureOne instead of using Lightroom or Photoshop CC as a RAW converter.

And I fully agree on your point that more resolution is good. Well - if you actually need it and have enough processing power to work with the files. If you don't need it the downsizes files will surely look better, no discussion here.
The only 'problem' I can see is that there's no guarantee the lens will deliver more resolution, I think a lot of lenses from the Hasselblad lineup will work but probably not all of them. As I said, right now you're getting the sweetspot of the center image - but on a 100mpx fullframe it might not look as good in the corners...

[eronald]

Erik,

 Thank you for making these interesting comparisons, and for posting them.

 I note the stitched Sony image of the city houses is higher resolving, exactly as you state - but esthetically I much prefer the P45+ image; the digital back is somehow much better regarding discrimination of the colors and especially the bricks.

 The more I see images from the Sony, the more I think it is an excellent camera in which some shortcuts were taken to raise ISO at the price of color.

Edmund

[ErikKaffehr]

Hi Edmund,

Thanks for comments, always appreciated!

I don't think it is about ISO. Anders Torger has looked into this and found that Kodak sensors like the one used in the P45+ have spectral characteristics that give very high colour separation but are very tricky to handle.

DALSA sensors have much more mainstream colour response curves. What I have seen the colour response curves on the Nikon D2X (using a Sony CCD sensor) where almost identical to the CMOS sensor used on the Sony NEX 5. Both were typical Sony sensors.

So, I think all vendors have a few mixes of dies used in different sensors. Kodak has extreme separation of reds. Canon has a lot of overlap between reds and greens.



Best regards
Erik

Eric,

 Thank you for making these comparisons.

 I note the stitched Sony image of the city houses is higher resolving - but esthetically I much prefer the P45+ image; the digital back is somehow much better regarding discrimination of the colors and especially the bricks.

 The more I see images from the Sony, the more I think it is an excellent camera in which some shortcuts were taken to raise ISO at the price of color.

Edmund

[eronald]

Erik,

 I have a problem, I can only switch on my "tech" brain cells under extreme pressure these days. So, I'm giving you my opinion on what I see -inferior color- but I haven't really switched on my brain. The curves you give are doubtless exact, and of course Andreas knows a lot about color software, but none of us really seems to have the final word on *why* various companies use one set of CFAs rather than another. The Hassy CEO I interviewed years ago was proud of his (Kodak?) sensors vs Canon, saying Canon's overlap (he called it small angle) was due to to a wish of ISO and led to bad color discrimination; I would agree with him, but add that by selecting a CFA, a manufacturer chooses whether they prefer chroma noise or luminance noise in the shadow tones. Think of this as a corollary of information theory.  It is also possible that certain type of in-camera processing and noise reduction -remember Jpegs, backs didn't do them- work better with one type of CFA and noise rather than another , thereby forcing the hand of the dSLR designer.  As an aside, I hated profiling dSLRs, they tended to chroma noise explosion in the shadows and the profiles needed tweaking to kill this. The signal processors used in-camera may also force certain choices with respect to normalisation and locality. I'm sure some designer who happens to stroll by could illuminate the issue for us. Although illuminators do have a rather bad rep in these days of laser-guided weaponry, but that's a story for another day.

Edmund

Hi Edmund,

Thanks for comments, always appreciated!

I don't think it is about ISO. Anders Torger has looked into this and found that Kodak sensors like the one used in the P45+ have spectral characteristics that give very high colour separation but are very tricky to handle.

DALSA sensors have much more mainstream colour response curves. What I have seen the colour response curves on the Nikon D2X (using a Sony CCD sensor) where almost identical to the CMOS sensor used on the Sony NEX 5. Both were typical Sony sensors.

So, I think all vendors have a few mixes of dies used in different sensors. Kodak has extreme separation of reds. Canon has a lot of overlap between reds and greens.



Best regards
Erik