Sensor vs Lens constraint

[pgk]

Advertisement (gone after registration)

A light meter.

OK:D I'll give you that. I'll add in...And the position of each photon - how accurately the position needs to be recorded of course can be debated here for a very long time....

[k-hawinkler]

Add in frequency or energy of photons as well?

[mjh]

Yes. A simple enough answer is that it would be a sensor which could record each and every individual photon arriving from the lens;). What this would achieve I have no idea....

What it would achieve is a sensor with no fixed ISO sensitivity, a potentially unlimited dynamic range, a variable trade-off between resolution and noise/dynamic range etc.. The photon counting sensor is something like the holy grail of sensor design. Some people believe we will live to see it.

[lars_bergquist]

To conclude, a lousy sensor and a lousy lens gives the best result

 

– when coupled to some processing wetware and software that makes the Macintosh and Photoshop look archaic.

 

The 20/20 old man

[giordano]

What it would achieve is a sensor with no fixed ISO sensitivity, a potentially unlimited dynamic range, a variable trade-off between resolution and noise/dynamic range etc.. The photon counting sensor is something like the holy grail of sensor design. Some people believe we will live to see it.

 

And if the sensor records every photon arriving not from the lens but from the scene, the result should be a Lytro light-field camera with all the above in addition.

[pgk]

To put things into a very simplistic analogy, if you photograph or scan an image in a newspaper at ever increasing resolutions, then there comes a point where you may well produce greater detail of the dots which go to make up the image, but are you really providing any additional information which is actually relevant to the image?

[haroldp]

Advertisement (gone after registration)

To put things into a very simplistic analogy, if you photograph or scan an image in a newspaper at ever increasing resolutions, then there comes a point where you may well produce greater detail of the dots which go to make up the image, but are you really providing any additional information which is actually relevant to the image?

 

The image projected by the lens onto the sensor is not discrete like halftone newsprint photos, but a set of declining conrast curve functions. This is what MTF tries to describe.

 

.... H

[01af]

To put things into a very simplistic analogy ...

Simplify things as much as possible—but not more!

[LucisPictor]

Calm down, guys, please.

 

I guess one of the reasons why the question has been asked are statements like: "The resolving power of the Zeiss Biogon 25mm f2.8 is 400 lp/mm!"

 

From there it is a short step to the calculation of how many MPix that would mean.

 

It is not the way it should be done, but it is often done.

[pgk]

The image projected by the lens onto the sensor is not discrete like halftone newsprint photos, but a set of declining contrast curve functions. This is what MTF tries to describe.

 

.... H

At high magnification the halftone newsprint isn't discrete either, but does the ability to show this have any relevance? My point was, and is, that just obtaining lots more data by over sampling is not going to provide more relevant data in the image. Determining at what point the excess data becomes meaningless is the trick. The problem we have is trying to quantify a very complex system in a simplistic way and as is often the case we try to provide some straightforward numbers to do this. You only have to look at the numerous 'tests' which come up with an overall 'score/mark/percentage' to realise how much we apparently love such simplistic descriptions.

[01af]

My point was, and is, that just obtaining lots more data by oversampling is not going to provide more relevant data in the image.

'Relevance' is a subjective category ... and 'oversampling' means that you don't fully understand the difference between your simplification and the real world. There is no oversampling in digital photography.

 

Of course, when you're able to agree upon some definition of what kind of resolution is relevant and what is not, then (and only then) you can introduce a limit for a useful pixel count ... which however will be higher than you probably think.

[pgk]

'Relevance' is a subjective category ... and 'oversampling' means that you don't fully understand the difference between your simplification and the real world. There is no oversampling in digital photography.

 

Of course, when you're able to agree upon some definition of what kind of resolution is relevant and what is not, then (and only then) you can introduce a limit for a useful pixel count ... which however will be higher than you probably think.

01af

 

In the real world we already substantially 'oversample' a most of the time. Existing cameras already have sensors capable of producing files far larger than the vast majority of photographs taken actually need, and are capable of quality levels that only a relatively small number of photographers actually utilise. My simplification was a simplification to illustrate that providing more data doesn't necessarily provide a better result.

 

As it happens my background is in scientific photography and I would welcome much higher senor resolution if it could yield more useful data. I have researched into the real world usage of digital cameras as a scientific too to aid identification. Unfortunately my findings are that whilst a doubling of MPixels does actually produce a little extra data, lens constraints start to kick in at 20MPixel+ and whilst there are theoretical increases in data, these can all too easily be offset by the physical costraints involved in taking the images (aperture requirements for example).

[adli]

Wow, finally someone who actually answers my question!

[lct]

...whilst a doubling of MPixels does actually produce a little extra data, lens constraints start to kick in at 20MPixel+...

Do you mean for any sensor size or for 24x36 only?

[leicafan!]

Oh man!

[leicafan!]

This thread has to have prompted the most outlandish conversations I've seen for a while!

[pgk]

Do you mean for any sensor size or for 24x36 only?

24x36 - larger cameras are too unwieldy for the application.

 

FWIW 35mm (Velvia 50) produced images which were about on a par with 10MPixel fimes for usable information (not in terms of aesthetics I hasten to add;)) and the difference between 5D and 5D2 files, whilst apparent was not as substantial as might be thought and little extra of the fine detail data required was being provided.

[haroldp]

24x36 - larger cameras are too unwieldy for the application.

 

FWIW 35mm (Velvia 50) produced images which were about on a par with 10MPixel fimes for usable information (not in terms of aesthetics I hasten to add;)) and the difference between 5D and 5D2 files, whilst apparent was not as substantial as might be thought and little extra of the fine detail data required was being provided.

 

I believe that much of the difference in the 'look' between Nikon D3 (12 mp) and D3x (24 mp), which is not subtle ( I can see it in a 4 X 6 ), is because the AA filter for the finer pixel pitch D3x is tuned to a higher nyquist frequency, thereby allowing the lens to be 'sharper'.

 

I have never used Canon DSLR's, but I recall they were taking some heat in the blogs around the time of changing to the 21 mp sensor, asserting that they kept the AA filters more aggressive ( fuzzier ) than they had to be with the new sensors, thereby blurring some of the benefit. I do not know how credible these blogs were.

 

In any case, most of the posts in this thread have been technically accurate, the question of what one is or should be willing to pay for any level of diminishing returns based not only on cost, but size, weight, inconvenience etc. is a personal one and cannot have a definitive answer.

 

Regards to all .... H

[lct]

...the difference between 5D and 5D2 files, whilst apparent was not as substantial as might be thought and little extra of the fine detail data required was being provided.

Interesting indeed. I did not update my 5D1 for this very reason and because i prefer its IQ with R lenses but it is another story.

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[philipus]

Unfortunately my findings are that whilst a doubling of MPixels does actually produce a little extra data, lens constraints start to kick in at 20MPixel+ and whilst there are theoretical increases in data, these can all too easily be offset by the physical costraints involved in taking the images (aperture requirements for example).

 

I find this very interesting. Could you elaborate a bit more on the particular lens constraints you've found?

 

Cheers

Philip