ErikKaffehr
Well-known member
Hi,
I would say that there are like three different factors:
- Sensor size
- Resolution
- Noise
A larger sensor will collect more light. Most of the noise in digital photography is a property of light (called shot noise). This is caused by the statistical distribution of light. A sensor that collects more light will have less shot noise.
So a sensor four times 24x36 will have twice the Signal/Noise Ratio. Now, if you reduce exposure two stops on the MF sensor, the Signal/Noise ratio will be the same.
Another factor is readout noise, that only affects the darks (as shot noise normally dominates). Readout noise is something like four times higher on CCD sensors compared with Sony CMOS (Nikon/Sony/Phase One IQ-250). This is the main reason CMOS is good at high ISO. You can underexpose more on CMOS because the signal is much cleaner. The technology used with CCDs on digital cameras doesn't allow low readout noise.
Resolution helps in detecting fine detail, but visual impression is dominated by medium size features. A rule of thumb may be that 20 lp/mm dominates visual expression at 24x36 mm. At full frame MF perhaps we could use 12 lp/mm. A good lens may yield 80% MTF at 20 lp/mm, some may go up to 90% MTF at 20 lp/mm (note that this figure is not possible at f/8, because of diffraction limit).
The same lens could achieve 90% at 12 lp/mm.
But, proper sharpening can kick MTF up to 100%. Sharpening is more important than lens quality, except when we pixel peep at 70 lp/mm, which is totally irrelevant in most contexts.
Now, a bad lens will show artefacts, like double contours, color fringing etc. It may be more important that a lens delivers a clean image than it delivering a very sharp image.
If the lens outresolves the sensor, aliasing will result. This may or may not be objectionable. Bad techniques, like bad focus, excessive stopping down etc can reduce aliasing. Also, aliasing will not arise without the presence of fine detail.
This posting shows the benefits of small pixels:
The Future of CCD Sensors
This videos from Panavision & Canon described the issues, from a Hollywood perspective:
https://www.youtube.com/watch?feature=player_detailpage&v=iBKDjLeNlsQ
https://www.youtube.com/watch?feature=player_detailpage&v=v96yhEr-DWM
It seems that in motion they want to avoid sharpening, they want 90% MTF at 20lp/mm in raw. That may be because sharpening may cause problems in postprocessing. I also guess that sharpening is less elaborate on a 2MP image at 24 frames/s than on a 39 MP image with individual processing.
Achieving 90% MTF at 20 lp/mm is very hard. Some lenses achieve it, Zeiss Otus 55/1.4 achieves it over the field at f/4 but the Zeiss Apo Sonnar 135/2 goes below 90% MTF at f/4 10 mm of axis.
The best of my MF lenses is probably the Sonnar 180/4 and that achieves a tiny bit more than 80% at f/8.
Best regards
Erik
I would say that there are like three different factors:
- Sensor size
- Resolution
- Noise
A larger sensor will collect more light. Most of the noise in digital photography is a property of light (called shot noise). This is caused by the statistical distribution of light. A sensor that collects more light will have less shot noise.
So a sensor four times 24x36 will have twice the Signal/Noise Ratio. Now, if you reduce exposure two stops on the MF sensor, the Signal/Noise ratio will be the same.
Another factor is readout noise, that only affects the darks (as shot noise normally dominates). Readout noise is something like four times higher on CCD sensors compared with Sony CMOS (Nikon/Sony/Phase One IQ-250). This is the main reason CMOS is good at high ISO. You can underexpose more on CMOS because the signal is much cleaner. The technology used with CCDs on digital cameras doesn't allow low readout noise.
Resolution helps in detecting fine detail, but visual impression is dominated by medium size features. A rule of thumb may be that 20 lp/mm dominates visual expression at 24x36 mm. At full frame MF perhaps we could use 12 lp/mm. A good lens may yield 80% MTF at 20 lp/mm, some may go up to 90% MTF at 20 lp/mm (note that this figure is not possible at f/8, because of diffraction limit).
The same lens could achieve 90% at 12 lp/mm.
But, proper sharpening can kick MTF up to 100%. Sharpening is more important than lens quality, except when we pixel peep at 70 lp/mm, which is totally irrelevant in most contexts.
Now, a bad lens will show artefacts, like double contours, color fringing etc. It may be more important that a lens delivers a clean image than it delivering a very sharp image.
If the lens outresolves the sensor, aliasing will result. This may or may not be objectionable. Bad techniques, like bad focus, excessive stopping down etc can reduce aliasing. Also, aliasing will not arise without the presence of fine detail.
This posting shows the benefits of small pixels:
The Future of CCD Sensors
This videos from Panavision & Canon described the issues, from a Hollywood perspective:
https://www.youtube.com/watch?feature=player_detailpage&v=iBKDjLeNlsQ
https://www.youtube.com/watch?feature=player_detailpage&v=v96yhEr-DWM
It seems that in motion they want to avoid sharpening, they want 90% MTF at 20lp/mm in raw. That may be because sharpening may cause problems in postprocessing. I also guess that sharpening is less elaborate on a 2MP image at 24 frames/s than on a 39 MP image with individual processing.
Achieving 90% MTF at 20 lp/mm is very hard. Some lenses achieve it, Zeiss Otus 55/1.4 achieves it over the field at f/4 but the Zeiss Apo Sonnar 135/2 goes below 90% MTF at f/4 10 mm of axis.
The best of my MF lenses is probably the Sonnar 180/4 and that achieves a tiny bit more than 80% at f/8.
Best regards
Erik
Thanks Erik, I think this is exactly what I was trying to remember and clarify. So if we look at the same print size the one made with larger format cameras will still have image quality benefit (MTF as you noted) and this is due to resolution benefit and optical system as Shlomi noted. So I was wondering if one can leverage this and get away with a less sharp lens since it kind of gains over a smaller format if one is willing to downsize the final images for use. Anyhow I ended up purchasing the D lenses which is a better decision in the long run but seems like it potentially could.
Is there any other benefit for a super high res sensor to leverage against if one is willing to downres? I happened to read this on Ming Thein's post this week about the Pentax 645.
"consider this: if at the pixel level it loses a stop or two in noise to the D4S, but has nearly four times the pixel count – downsampling is going to yield an amazingly clean image, regardless of the ISO used, with that medium format look*. And that makes things interesting.
*Related to the depth of field properties of the actual focal length of the lenses, and irrespective of the field of view. Smaller formats mean shorter focal lengths for the same equivalent FOV, and the attendant depth of field properties that go with it – i.e. a lot."
So I see the benefit of noise masking, anything else? Does it mask camera shake or out of focus shots?
Yes sharpening is a different topic altogether, appropriate to final size and medium indeed.
Last edited: