Cameras compensating for an undesirable phenomenon

A recent article on Luminous Landscape:

http://www.luminous-landscape.com/essays/an_open_letter_to_the_major_camera_manufacturers.shtml

discusses a recent report from DXO labs that reveals how certain digital cameras apparently ameliorate an undesirable optical-photometric phenomenon. The author suggests that camera manufacturers either should not do that, or at least should "tell us" about it.

Here is my take on this situation.

**************

The basic technical story

a. As we increase the relative aperture (that's what the f-number tells us), the photometric exposure on the film or overall sensor surface changes in a predictable way.

A change in f-number from f/4 to f/1.4 increases the photometric exposure by almost exactly 8 times. This will be true for digital sensors as well as film.

b. In a digital camera, the "photodetectors at the bottom of chimneys" effect means that the light from a point on the object that passes through the outer regions of the aperture and lands on the sensor surface (at an oblique angle) is relatively less effective on the sensor than the same amount of light per unit surface area arriving via the central part of the aperture (at a less oblique angle).

c. Thus the impact on the sensor is no longer consistent with photometric exposure, a discrepancy that escalates when more of the light arrives at more oblique angles, as occurs as the aperture increases.

Thus, perhaps, a change in the f-number from f/4 to f/1.4 would increase the impact on the photodetectors by only about 6 times.

d. As a result, if we use shutter speed priority and

• set a shutter speed of 1/200 sec, and the metering system set an aperture of f/4, and we got an exposure result that we thought was "proper", and then

• set the shutter speed to 1/25 sec, and the metering system would then set an aperture of f/1.4, we might find that we had a result that was underexposed by the equivalent of 1/3 stop.

e. Evidently, to avert this effect, the camera manufacturers shift the sensor gain up a bit when a larger aperture is about to be used.

f. The charts in the article suggest that, for a Canon 550D, when the aperture to be used is f/1.2, the gain is bumped up by about 1/2 stop (that is, by a factor of about 1.4).

We are tempted to say that this is an increase in the ISO SOS (an actual objective measure of the sensor sensitivity) of 1/2 stop (1.4x). But, as we will see shortly, that is not the proper outlook.

For a Nikon D200, when an aperture of f/1.4 is about to be be used, the sensor gain is evidently bumped by about 1/3 stop (a factor of about 1.25).

g. The objective measures of sensor sensitivity stated as various "ISO" quantities (ISO speed, ISO SOS) are based on the response of the sensor to various photometric exposures. The phenomenon at issue here reduces the sensitivity of the sensor as a greater fraction of the arriving light arrives at increasingly oblique angles.

Thus, the increase in gain said to be applied in such cases does not increase the ISO sensitivity but rather partially restores it to the "set" value. (The compensation is apparently not complete.)

Complain! Sue!

Great outrage over this apparent camera feature is suggested. But over what? That the manufacturer causes the camera to unilaterally increase the ISO sensitivity of the sensor to compensate for the expected phenomenon when a large aperture is about to be used? Well, they don't increase the ISO sensitivity. They do increase the sensor gain to prevent the decline of the ISO sensitivity.

In any event, if we for some reason don't like that, what do we suggest the manufacturers do instead?

The phenomenon is present (although lenses designed for digital cameras often have their entrance pupils intentionally moved forward to reduce the magnitude of the phenomenon). We cannot complain it out of existence.

Possibilities:

1. Do nothing. People might want to employ exposure compensation (EC) to compensate for the metering inconsistency with large apertures. Or complain. Horn players have to move their fists inside the bells of their instruments to keep them properly on pitch for different notes. Singers move a hand microphone farther from their mouths when singing louder. It's part of the craft.

2. Automate the "application of EC". In aperture priority mode, set a shutter speed that differed from the basic result of the metering equations to compensate for the phenomenon expected at the aperture about to be used. In shutter speed priority mode, set an aperture that would take the phenomenon into account. In programmed exposure mode, set a combination that would take the phenomenon into effect. In manual exposure mode, make an adjustment to the "little meter".

A disadvantage is that the exposure set might be inconsistent with the indications of an external exposure meter (something that camera manufacturers strive mightily to avoid, since it leads to complaints about the camera's metering system being inaccurate).

3. Adjust the sensor sensitivity to compensate (apparently what is being done).

If the complaint is that the manufacturers don't tell us this, I could fill pages with things they don't tell us. Canon, for example, doesn't tell us that with a flash active, and the camera in programmed exposure mode, an aperture larger than f/4.0 will never be selected. Or how E-TTL metering works.

The argument is made that by doing this, the manufacturers are misrepresenting the "ISO something" of their cameras under this situation. Of course, as I pointed out, that is not really what happens.

In any case, note that Canon, for example, does not say that the "ISO" setting we make on the camera produces a corresponding value either of the two ISO objective measures of actual sensitivity, the ISO speed or the ISO SOS. They state instead the ISO REI, a value they are free to choose as they see fit.

Loss of light? Or just squandering it.

The author speaks of the underlying phenomenon as being a loss of light. It isn't so far as the overall sensor is concerned, although it is from the standpoint of the business end of individual detectors themselves.

What does the concept of the T-stop have to do with this?

The author also characterizes this phenomenon as a degradation in the T-stop of the lens. As he well discusses at the outset, the T-stop of a lens (for any given f-number setting) is sort of an "effective f-number" which takes into account that the lens does not have 100% transmission - that is, that part of the light it collects is not delivered to the focal plane.

But it is misleading to invoke this as a metaphor in this situation. The phenomenon at issue does not cause the lens to deliver less light to the focal plane. It does cause the individual photodetectors to catch and use less of that light when the light arrives at increasingly oblique angles.

Another aspect of the matter

The phenomenon of the sensor having less sensitivity in the case of "oblique rays" has another bad effect, not so correctable by a change in exposure or in sensor gain.

For a given scene point luminance, the photometric exposure falls off as we move from the center of the scene (the source of "natural vignetting"). One classical model of this phenomenon suggests that the falloff is proportional to the fourth power of the cosine of the angle the point of interest lies off the axis. Of course, this is rarely found to be the actual relationship.

But, whatever the relationship is for a particular lens at a particular aperture, at larger apertures, the phenomenon we discuss here (the "photodetectors at the bottom of chimneys" effect) accelerates the decline not in photometric exposure but rather in the impact on the sensor (which is of course what creates our image).

One can of course compensate for the overall effect during raw development (or later, if need be). Many processing software packages do this, based on tables for the lens noted to be in use), and some cameras are now beginning to have this internally.

Well, it's time for breakfast.

Best regards,

Doug

Doug,

Obfuscation! Yes, one can parse a definition of ISO legally, (and logically), to a form with wiggle room for MFRS is beside the point. The issue is simply this. The use of peripheral light creates mood and imaging effects that are predicted from experience with film, where, for all intents and purposes, unique incident light focus and flux geographical differences are reflected in the patterns and ~ proportional silver grain deposition in subsequent film development. It's the nature of the image at f 1.2 not the exposure that's the main point. The gain is a secondary issue.

So photographers who desire the unique local and general effects of light coming from the edge of the lens can dial in such artistic nuances by increasing aperture and using more of the edges of that lens. What the MFRS have not disclosed is that even with the latest post DSLR era modern f 1.2 lenses, digital sensors cannot record the peripheral light to the extent film does and so they boost overall gain to disguise that fact. IOW, the f1.2 lens might really be more like an f 1.4 lens as an artistic tool.

So folk like me who paid a premium for a modern 50 1.2L lens or an 85 1.2L lens are not likely to get the effect they's expect from using film, even after the gain is pumped up. After all, it's just the light from the periphery that needs to be corrected for.

Also, since noise increases with boosting gain, we'd like to know that. Not clearly disclosing this is plain wrong.

Asher

Doug,

Let me add that in selling powerful medicines, as new information comes to the MFR's attention, they are required by law to disclose this. For example, Avastin is not to be used for such as such cancer or in case of this bacterial infection in such and such population, penicillin-resistance should be considered.

When the mechanical design of the sensor cannot use the optical output of the lens as traditionally expected with film, then that should be disclosed.

Asher

Asher Kelman said:

It's the nature of the image at f 1.2 not the exposure that's the main point. The gain is a secondary issue.

Click to expand...

Do we have any idea how much difference in the nature of the image (bokeh, etc.) we get between a film and digital image, at the same aperture, for large apertures, as a result of this phenomenon?

I would hate to rush to the judgment that there is any substantial (even perceptible) change.

A terribly simplistic model is that all the light that would come through the pupils of an f/1.2 lens beyond the pupil size of an f/1.4 lens is ineffectual on out digital sensor (an in-or-out approach, obviously not realistic).

If we then look at the diameter of the circle of confusion (not the circle of confusion diameter limit we use to reckon depth of field - the actual diameter of the circle of confusion that is caused by a specific misfocus situation, the stuff of bokeh), for this case:


then for an f/1.2 lens with no "peripheral ineffectiveness", the diameter of the circle of confusion would be 0.41 mm.

For an f/1.2 lens with all the light beyond the f/1.4 pupil diameter ineffectual, the diameter of the circle of confusion would be 0.35 mm.

This primitive model may of course not give a meaningful suggestion of the actual difference in behavior. But it is interesting.

And remember, in the Canon EF lenses, if you choose to spend the money for the 50 mm f/1.2 rather than the 50 mm f/1.4, you get a much nicer full time manual focusing mechanism.

Best regards,

Doug

Hi, Asher,

Asher Kelman said:

When the mechanical design of the sensor cannot use the optical output of the lens as traditionally expected with film, then that should be disclosed.

Click to expand...

Well, that makes sense.

Note however that this fact is well known in the digital camera world and has been for some while. It is of course not in the camera catalog pages (like many other critical things).

Best regards,

Doug

Given that each manufacturers definition of the iso varies even between camera ranges and the biggest differences were shown at under half a stop in a 5 year old camera (5D) I really think that this is rather yawn making. A bit too anal for photographers given how good the IQ is and how little of this iso pushing is happening with current models. I'm still shooting with 5D's and I'd glad that the camera is pushing the iso rather than me having to worry about tstops, especially given how clean this sensor is anyway. Problem is that unlike as this article suggests, it doesn't. Shooting using studio lighting changing from a 50mm prime to a zoom will lose me 2/3 of a stop due to the difference in Tstop value and it's very noticeable. It shouldn't be if this article was right but I see it often..

Ben Rubinstein said:

Given that each manufacturers definition of the iso varies even between camera ranges and the biggest differences were shown at under half a stop in a 5 year old camera (5D) I really think that this is rather yawn making. A bit too anal for photographers given how good the IQ is and how little of this iso pushing is happening with current models. I'm still shooting with 5D's and I'd glad that the camera is pushing the iso rather than me having to worry about tstops, especially given how clean this sensor is anyway. Problem is that unlike as this article suggests, it doesn't. Shooting using studio lighting changing from a 50mm prime to a zoom will lose me 2/3 of a stop due to the difference in Tstop value and it's very noticeable. It shouldn't be if this article was right but I see it often..

Click to expand...

Ben,

The issue is not really with the fiddle but rather than the loss of the specific portion of light from the edges of the f 1.2 L lens. That's what gives part of the nature of the f 1.2 in film. That portion is simply lost by boosting overall the light arriving on the sensors.

They should state that this lens cannot reproduce the full open lens effects seen on film camera when using digital. As for gain, that's a red herring although should be disclosed. Of course, I'll have to do a study now to see if the effects are indeed discernible and perhaps this is not a practical issue, LOL!

Asher

Hi, Ben,

Ben Rubinstein said:

Shooting using studio lighting changing from a 50mm prime to a zoom will lose me 2/3 of a stop due to the difference in Tstop value and it's very noticeable. It shouldn't be if this article was right but I see it often..

Click to expand...

Nothing in the article runs contrary to your experience. The zooms no doubt have lower transmission than the primes and thus a larger T-stop for the same f-number, this having nothing to do with marginal rays. The article doesn't deal with this at all (which is why even mentioning T-stops is a red herring).

Best regards,

Doug

Hi, Asher,

Asher Kelman said:

That portion is simply lost by boosting overall the light arriving on the sensors.

Click to expand...

I don't know what you mean "boosting the overall light arriving on the sensors". And what could we not do so as to cause that light not to be lost?

Best regards,

Doug

The DxO data cited in the Luminous Landscape article suggests that at an aperture of f/1.2, the "chimney phenomenon" is over twice as severe on an EOS 7D as on an EOS 5D. That is probably as great a difference as we can expect to observe with familiar stuff.

Is anyone in a position to conduct a test with those two bodies, both with an EF 50 mm f/1.2 lens at f/1.2, with the same scene(s) (a tripod lockdown would be lovely), one(s) that would allow us to observe the nature of the bokeh?

We would want to compare the full-frame output of the 7D against a 2720 x 1815 px crop from the 5D (both 22.3 x 14.9 mm at the focal plane).

If somebody would like to ship me a 5D, a 7D, and a 50/1.2, I'd be glad to do it here. I would bear the cost of shipping the stuff back - no stuff from Botswana, please.

Or if somebody wants to try it at f/1.4 (where the difference between bodies would be not quite so great, but might be more representative of wider usage), I'd be glad to loan my 50/1.4 into the project.

Best regards,

Doug

Doug Kerr said:

Hi, Asher,



I don't know what you mean "boosting the overall light arriving on the sensors". And what could we not do so as to cause that light not to be lost?

Best regards,

Doug

Click to expand...

Read that as "boosting the electronic response to light received all over the CMOS sensor". The fraction of light coming from the periphery, (having an artistic impact on the resultant picture) is this made smaller.

Asher

Hi, Asher,

Asher Kelman said:

Read that as "boosting the electronic response to light received all over the CMOS sensor".

Click to expand...

I understand.

The fraction of light coming from the periphery, (having an artistic impact on the resultant picture) is this made smaller.

Click to expand...

No, increasing the gain does not change the relative contributions of central and peripheral light. That is only a function of the degree of the "chimney effect", whatever the gain.

Changing the gain will have the same effect, in this regard, as changing the shutter time.

Of course it is always possible that with an overall "underexposure" (as if we do not jack up the gain), some of the artistic effects might look better, but that is a different matter altogether.

Best regards,

Doug

Doug,

I'm only considering the end result. By not making up the fraction of the lost light due to the chimney effect, for that component only, the light from the edge of the lens is not fully represented in the light recorded. Then when Canon adds gain to the entire sensor to make up that loss, the fraction coming from the periphery is thus decreased compared to the fraction that would be recorded on film.

asher

Hi, Asher,

Asher Kelman said:

I'm only considering the end result. By not making up the fraction of the lost light due to the chimney effect, for that component only, the light from the edge of the lens is not fully represented in the light recorded. Then when Canon adds gain to the entire sensor to make up that loss, the fraction coming from the periphery is thus decreased compared to the fraction that would be recorded on film.

Click to expand...

Imagine two regions across the exit pupil, a small circular one at the center, and a thin annular one at the periphery, having such areas that the light through both makes equal contributions to the illuminance on the focal plane.

With film, both will then make equal contributors to the latent photochemical exposure result.

But suppose that, on a certain digital sensor, owing to the chimney effect, the peripheral one makes half the contribution to the digital exposure result as does the center one. We will arbitrarily say that these are numerically 10 and 20 units respectively - a ratio of 1 to 2, peripheral to central.

Also suppose that the the total digital exposure result is 0.67 times what it would be if there were no chimney effect (not all the light is discounted to 1/2 of its impact - the light through the central region is not discounted at all). Suppose then that in this situation we therefore bump the gain by a factor of 1.5 to restore the "proper" digital exposure result.

Then the digital exposure impact of the light from the peripheral annulus would be 15 units (1.5 x 10), and the digital exposure impact of the light from the central region would be 30 units (1.5 x 20) - a ratio again of 1 to 2, peripheral to central.

So the relative contributions of the light from the peripheral and central regions is unchanged by a change in gain.

The ratio of the contributions to exposure effect of the light from these two regions is of course different in the digital case from what it would be with film, because of the chimney effect, but that ratio of contributions is not further changed by a change in gain.

Best regards

Doug

Doug Kerr said:

Hi, Asher,



Imagine two regions across the exit pupil, a small circular one at the center, and a thin annular one at the periphery, having such areas that the light through both makes equal contributions to the illuminance on the focal plane.

With film, both will then make equal contributors to the latent photochemical exposure result.

But suppose that, on a certain digital sensor, owing to the chimney effect, the peripheral one makes half the contribution to the digital exposure result as does the center one. We will arbitrarily say that these are numerically 10 and 20 units respectively - a ratio of 1 to 2, peripheral to central.

Also suppose that the the total digital exposure result is 0.67 times what it would be if there were no chimney effect (not all the light is discounted to 1/2 of its impact - the light through the central region is not discounted at all). Suppose then that in this situation we therefore bump the gain by a factor of 1.5 to restore the "proper" digital exposure result.

Then the digital exposure impact of the light from the peripheral annulus would be 15 units (1.5 x 10), and the digital exposure impact of the light from the central region would be 30 units (1.5 x 20) - a ratio again of 1 to 2, peripheral to central.

So the relative contributions of the light from the peripheral and central regions is unchanged by a change in gain.

Click to expand...

That's my point! The change in gain does not make up for the deficit in the peripheral zone!@ Wghy you have to calculate to get that I don't know. It's just obvious!

Doug Kerr said:

The ratio of the contributions to exposure effect of the light from these two regions is of course different in the digital case from what it would be with film, because of the chimney effect,

Click to expand...

Hello! That's my point! We pay for the f 1.2 L aperture and the artistic effect it can yield. If, indeed, it turns out, to be a real loss of the recording of the unique quality of light from the lens periphery, (detectable in comparative tests, film v. digital), as a deficiency in the lens as an photographic painting tool, I'll be mad at them. They didn't tell us about this. But we have to demonstrate that this theoretical loss is indeed something substantial.

Doug Kerr said:

but that ratio of contributions is not further changed by a change in gain.

Click to expand...

Of course! Who cares about the electronic gain, it cannot bring back a unique fraction from the edge of the lens if that's lost and if the consequence is significant!

Asher

Doug Kerr said:

Best regards

Doug

Click to expand...

Equally my friend!

Hi, Asher,

Asher Kelman said:

That's my point! The change in gain does not make up for the deficit in the peripheral zone!

Click to expand...

Well said.

Best regards,

Doug

I'm confused. The 'look' of the 85L is still utterly unique, on digital or not. The idea that the DOF is changed due to this phenomenon is still to be substantiated.

Ben Rubinstein said:

I'm confused. The 'look' of the 85L is still utterly unique, on digital or not. The idea that the DOF is changed due to this phenomenon is still to be substantiated.

Click to expand...

Ben,

I can only test the 50 1.2 L, the 50 i.4L and the 50 1.8L but not the 85mm as I don't own one. Right now my 50 1.2L has to go into Canon CPS as the front unit of the lens has come loose again!

Asher

Hi, Ben,

Ben Rubinstein said:

I'm confused. The 'look' of the 85L is still utterly unique, on digital or not. The idea that the DOF is changed due to this phenomenon is still to be substantiated.

Click to expand...

There is no question that the DoF, and the out of focus blur performance (which, while related, are not the same thing) should be affected by this phenomenon.

But whether the impact on the image is "perceptable", or "consequential", indeed remains to be seen.

I'm still waiting for the needed apparatus to arrive here so I can conduct tests in this regard.

Best regards,

Doug

Doug, that's interesting! What are you planning?

Asher

Hi, Asher,

Asher Kelman said:

I can only test the 50 1.2 L, the 50 i.4L and the 50 1.8L but not the 85mm as I don't own one. Right now my 50 1.2L has to go into Canon CPS as the front unit of the lens has come loose again!

Click to expand...

I didn't know that there was a 50/1.4L and a 50/1.8L. Is the 50/1.4L ring USM?

It would probably be worth testing with the 50/1.4. What bodies might you do that on? The best pair would be ones showing the greatest difference in "delta EV" on this chart:


Best regards,

Doug

Doug,

it would make sense to use the 5D and 5D Mark II. I have a 1DII but it's not full frame. I don't know how to figure that one out as one loses the outer perimeter anyway.

What test set up would you suggest?

Asher

Hi Asher,

Asher Kelman said:

Doug,

it would make sense to use the 5D and 5D Mark II. I have a 1DII but it's not full frame. I don't know how to figure that one out as one loses the outer perimeter anyway.

Click to expand...

First note that this is a different meaning of "perimeter". I have to trouble you with some theory here, as there is a real chance of misunderstanding between two quite different concepts. It is best to get this out of the way here.

*********

When we use a camera that has a smaller frame size than the lens was designed to support, it does not use all of the lens' image circle. This has nothing to do with how big a hole does the light come through, but rather how far off the axis can an object be and still have its image fall in the frame.

Typically lens performance falls off in various ways for object at greater distances off the axis. Thus, with a smaller frame camera, we often are happy to no longer deal with this less satisfactory aspect of the lens: how it deals with objects far off the axis - objects that are far off the axis are no longer "in the picture" (in both literal and figurative senses).

For any object point any place within the field of view, light enters through every part of the entrance pupil, emerges through every part of the exit pupil, all falling onto the focal plane at the same place, forming a point of the image.

The issue in the current topic is wholly different. Start with the blue passage above.

As we reduce the aperture, the size of the entrance pupil decreases (the exit pupil with it). Thus the light that, for a larger aperture lens, came through the outer reaches of the pupils is no longer collected, nor delivered to the image point. The result of course is that the illuminance on the focal plane, at that point, is reduced, with a corresponding effect on photometric exposure.

Now, when the chimney effect comes into play, from regard to any object point within the field of view, light that is indeed collected by the outer reaches of the entrance pupil, and is indeed delivered out of the outer reaches of the exit pupil onto the focal plane at the image point, is nevertheless "discounted" in its effect on the sensor. It arrives safely, but doesn't do as much as the same amount of light passing through the center of the pupils (maybe doesn't do anything).

Just for reference, the formal name for the light that passes through the outermost part of the pupils is the marginal rays.

*********

What test set up would you suggest?

Click to expand...

The difference in "chimney effect" is evidently not as great between the 5D and the 5D2 as we might like, but it is greater than between either of those and the iD2, so probably the two 5's would be your best bet.

• I suggest you select a nice scene in which we might be concerned with the bokeh, with some object fairly near (perhaps a garden gnome), and other objects (little white flowers, perhaps) at a greater distance (maybe some nearer as well).

• Put body A on a tripod (quick release plates would be nice for consistency in aim), set the aperture to full (in aperture priority), focus on the garden gnome, and take a metered shot.

• Then put the lens on body B, put that body on the tripod, and do the same thing.

• Look at these two images at a comparable physical size on the monitor and see if you can perceive any difference in the nature of the bokeh.

It is possible that the difference in resolution between the two bodies could add a complication. Perhaps the 5D2 image should be downsampled to attain the same pixel dimensions as the 5D image.

Best regards,

Doug