ISO/shutter speed inconsistency

[Ray]

BJL,
Sorry! I missed this thread so ignore my comment in the F828 thread re ISO.

It seems to me that the maximum DR is going to be limited by the size of the photodetector even if it were possible to eliminate all other sources of noise and interference. However, there are certain types of noise that are a constant and these would be comparatively less significant in relation to the increased number of photons that result from a low ISO setting.

Trouble is, low ISO settings are not popular so Sony has thought fit to fib a little. It looks as though the F828 is an ISO 25 camera and it has to be this way in order to get reasonable shadow detail. I suspect all digicams would show less noise if they were optimised for a lower ISO rating.

[Ray]

Please cite some evidence for your claim that ISO 64 on the Sony is equivalent to ISO 25.

Matthew,
Notice I wrote, it looks as though the F828 is an ISO 25 camera. I haven't done any conclusive tests. I'm going mainly by what Fabio Riccardi has written in the other thread - ie. the F828 is at least one stop slower than the 10D. One stop slower turns ISO 64 into ISO 32. A bit more and you arrive at ISO 25.

However, I see some complication here in getting at the truth. I never got the impression my D60 was giving me wildly different exposures at 100 ISO than I was used to with my film based 35mm cameras, so I never bothered doing a comparison.

Just out of curiosity, I've just done a few tests with a Minolta Dynax 404si 35mm film camera. Fitted a 50mm lens to my D60, a 35-105 zoom (set at 80mm) to the Minolta and pointed both cameras in the same direction, ISO setting 100, aperture priority.

Interestingly, some of the time the exposure readings for both cameras were the same and some of the time the D60 showed the next faster shutter speed. For example, the Minolta might show 1/45th whereas the D60 would show 1/60th, but never the other way round. However, in such circumstances, changing the aperture in both cameras by 1/2 a stop would bring the exposures back in line. That seems to imply to me a different approach to rounding up or rounding down. Ie. if the 'correct' exposure is actually about 1/52 sec, the camera has to choose between 1/45th or 1/60th.

Another complication is the 'accuracy' of film ISO ratings. I've often come across the comment that Fuji's Velvia 50 is best used at an ISO 40 setting (or even less). On the other hand, negative film with its higher DR can often be safely overexposed by a whole stop (or more) in order to get better detail in the shadows, and without blowing the highlights.

In order to arrive at any firm conclusion about this, one would really have to shoot the pictures (same aperture and exposure) and see the results. I'm not in a position to do this with the F828.

[BJL]

Digging through some earlier reviews at DPreview (the Minolta A1?) Phil Askey indicates that the metering of the Sony 717 is correct (he checked absolute luminance levels), so that the 828 is giving exposures 1/3 stop slower and the 300D (and D60,10D?) aer giving exposures 1/3 stop faster than "standard".

   If one interprets this as an error in the stated speed rating, the 828 has a true minimum speed of 50 instead of 64, while the D60/10D/300D have a true minimum of 125. To be gratuitously cynical, maybe digicam makers tend to exagerate speed ratings, while DSLR makers sometimes pretend to have a lower minimum speed, since this is sometimes taken as an indication of greater highlight handling and dynamic range!

   However, as I have said before, it could also be that the speed ratings are correct, but the metering calibration differs due to something akin to the "K factor" that Ansel Adams complained about. That is, the meters are recommending a bit of under- or over- exposure according to what the maker thinks will give the best balance of highlight headroom and shadow handling for the particular sensor. (For comparison, Velvia apparently is a true ISO 50, but many rate it at 40, which is really choosing to overexpose slightly.)

    This cannot easily be tested by examining JPEG output, as that is transformed with tone curves, gamma compression and such that could hide the differences, so one way I can think of to test this is to take raw images of a uniform subject like a gray card, convert them linear (no gamma or tone curve or black point and white point adjustment ... ) and check what level the images come out at. (An ISO standard suggests that it should be 18/170 of maximum.)

HOWEVER,
   whatever the explanation, one can still compare noise curves this way: compute what shutter speeds give the same noise level for the same aperture and lighting conditions.

   With the 828 to 300D comparison, the 828 curve at DPReview is already about three stops to the left (noisier), and correcting for the 2/3 stop difference in exposure times, the gap is really more like 3 2/3 stop, or a factor of about 10 in shutter speed needed to get a certain noise level. For example, the 828 at f/5.6 and shutter speed of 1/100 has noise levels comparable to the 300D at f/5.6 and 1/1000 (and the Olympus E-1 at about f/5.6 and 1/500). The 828's lens speed advantage over the 300D kit lens and greater DOF at a given f-stop can recover about two or 2 1/3 stops of this under many circumstances, but it still falls about 1 1/3 stops or more behind the 300D (similar when compared to the E-1.)

P.S. As another example, my Olympus C-2040 meters one stop slower than any of my SLR's; that is, it recommends an exposure twice as long.

[Bobtrips]

I would say, on average, in sunny conditions, I'm using the same shutter speed on the D60 set at ISO 100, as I would have previously used on a film based camera set at ISO 250-300. Ie., I'm underexposing by a whole stop or more to preserve the highlight detail.

Are you finding that you need to underexpose that much?  

I've been using 0.3 bracketing and wishing I could set my camera auto-bracket for -0.3 and -0.6.  I've never found the +0.3 useful.

I've even tried 'cheating' by setting the camera to -0.3 and then I get 0, -0.3, and -0.6 shots when I hold the button for the full bracket set.  (I do a lot of 'quick shots' - 'street photography'.  There's often no time to think things through.)

[MatthewCromer]

No, but since the 828 meters very accurately and doesn't tend to blow highlights and has a live histogram, I don't need to set exposure comp way down and pull the exposure out of the shadows.

[BJL]

I think the issue of whether or not the ISO setting on a digital camera is accurate or not is quite trivial compared to the dynamic range issue.

Highly agreed about DR; and I have started to think that digital cameras should be described by a range of fully satisfactory speeds (Exposure Indices) rather than a single one, with a wider range for cameras with more dynamic range.

  The upper speed limit is set by the minimum exposure level needed to get adequate shadow handling, as is the case with film speed rating; the lower limit is set by the maximum exposure level that allows adequate highlight headroom, as with the standard for the "base ISO speed" of an electronic sensor.

   For the example I know the most technical data for, the  Olympus E-1, the range seems to be from about EI100 up to somewhere between EI400 and EI800; for the Canon 10D/300D, my guess is about EI125 to between EI800 or EI1600.

   Note that some DSLR's have a "good" speed range, but also options to use higher speeds (maybe up to EI3200, a kind of push processing where one must tolerate more visible noise) and/or lower speeds (usually just EI50, a kind of pull processing where one probably risks blown highlights with some subjects.)

    For the Sony 828, it seems that the lower speed limit is about EI50. The upper speed limit seems to be somewhere between EI100 and EI400; I should be vague since I have not seen much discussion or measurement of shadow handling as revealed by prints rather, but only statistical analysis of pixel values.

   I suppose that even good EI100 combined with the 828's high resolution and relatively low price is already quite attractive for certain styles of photography.

[Ray]

I shot a bunch of images of landscapes that included patches of snow and not one blew out the whites in the snow, unless I applied positive exposure compensation.

Matthew,
Excellent result! Sounds like a great little camera! Go for it! Don't let me try to spoil your fun.  

[BJL]

I have decided to salvage this issue from the now very long thread that started with a report of inflated ISO ratings of the Sony 828.

    What is reported there and in many other cases seems to be that with different digital cameras, you can set the same ISO and aperture ratio in aperture priority mode, point them at the same, fairly uniformly lit subject, and yet get a wide range of different shutter speeds. For example, I have read claims that Canon compact digicams tend to give about twice the shutter speed of others at the same ISO setting, and that the Olympus E-1 gives almost four times the shutter speed as Nikon SLR's (film and digital), in which case each is in some sense operating "faster" than their ISO setting indicates, while the Sony 828 reportedly goes in the opposite direction.

    Does anyone have solid information on this, and if it is true, what it could mean? Awaiting replies, here is some speculation.

   Choice of exposure time must trade off between the risks of blown-out highlights and visible noise/grain in shadows, and photographers often make different shutter speed choices with different films at the same ISO, aperture and light conditions. If one digital camera has better dynamic range than another (more stops between sensor illumination levels that blow-out and those that look poor due to dark current noise), it might make sense to use a lower exposure level, in the sense of placing objects of measured average brightness further down from the maximum signal level the sensor can handle; this would gain highlight headroom but move shadows closer to the dark noise level, relying on the better noise behaviour.

   (Conversely, I have read from Kodak that compact digicams with their far smaller pixels and so less dynamic range tend to set exposure levels higher than DSLR's, limiting the highlight headroom for high contrast subjects in order to keep the visibility of shadow noise down; this could translate to slower shutter speeds at a given ISO rating for digicams compared to DSLR's.)

[MatthewCromer]

Ray,

Please cite some evidence for your claim that ISO 64 on the Sony is equivalent to ISO 25.  That's really quite a claim and I think you should back it up.

At STF folks have tested their 717s and 828s and found that the ISO exposure matches what light meters recommend for film at the same ISO value.  So I'm very curious at the evidence that Sony overrates the ISO on their cameras (Phil Askey claims that Canon UNDERRATES their ISO values by a stop).

[Ray]

By now you will no doubt be aware that dpreview has a full review of the F828. On the ISO issue, here's a quote from the review.

"In our tests the DSC-F828 was approximately a third of a stop (0.3 EV) less sensitive than the DSC-F717 at the same selected ISO and approximately two thirds of a stop (0.7 EV) less sensitive than the Canon EOS-300D (Digital Rebel).
Thus ISO 200 on the DSC-F828 would be the same as ISO 160 on the Sony DSC-F717 (but that setting doesn't exist) or ISO 125 on the Canon EOS-300D."

Now, I know I'm making some assumptions here which might not be correct, but it's all I've got to go on. Assuming the Minolta Dynax film camera is giving a correct reading, the most likely interpretation (after re-thinking) is that the D60 is slightly underrating the ISO, ie. a true ISO 100 is really ISO 125 on the D60.

Assuming again that the D60 ISO rating is identical to the 300D ISO rating, then ISO 200 on the F828, equal to ISO 125 on the 300D and D60, is equal to ISO 100 on my film based Minolta Dynax. Ie, there's a full stop difference.

The F828 therefore becomes a true ISO 32 camera. Happy? Or do you disagree with my logic, or the emprical evidence?

Not specific to the thread, but interesting so I thought I'd mention it, is the way the CA virtually disappears at F8. Unfortunately, the aluminium foil tests do not give a clear indication of resolution at f8. Off the top of my head, is this not equivalent to f22 on a 35mm camera? Yet the resolution appears to be greater, but I'm not sure if this is just greater noise giving the illusion of greater resolution.

[Ray]

Digging through some earlier reviews at DPreview (the Minolta A1?) Phil Askey indicates that the metering of the Sony 717 is correct (he checked absolute luminance levels), so that the 828 is giving exposures 1/3 stop slower and the 300D (and D60,10D?) aer giving exposures 1/3 stop faster than "standard".

   If one interprets this as an error in the stated speed rating, the 828 has a true minimum speed of 50 instead of 64, while the D60/10D/300D have a true minimum of 125. To be gratuitously cynical, maybe digicam makers tend to exagerate speed ratings, while DSLR makers sometimes pretend to have a lower minimum speed, since this is sometimes taken as an indication of greater highlight handling and dynamic range!

  

I actually wondered after I'd written my previous post if I'd made an error in reasoning (perhaps I shouldn't have had that glass of wine   ) . I see I've been piling on the 1/3 stop increments when I should have subtracted one of them.

When my D60 shows ISO 100, it should being showing ISO 125. The F828 at ISO 200 is using the same shutter speed as a 300D set at ISO 125, but ISO 125 on a 300D is equivalent to ISO 160 on a film based camera or on the F717.

I therefore agree with your interpretation that the F828 could be viewed as having a true base speed of ISO 50. (Apologies to Matthew!   )

Your point about the 'K factor' seems to me to be even more relevant with current DSLRs than it was in the days of Ansel Adams. I'll always remember those ruined shots with blown highlights from the first batch I took with my new D60, shot with a similar evaluative metering method to what I'd been using previously with my film based camera. And I didn't even save the RAW images  .

I would say, on average, in sunny conditions, I'm using the same shutter speed on the D60 set at ISO 100, as I would have previously used on a film based camera set at ISO 250-300. Ie., I'm underesposing by a whole stop or more to preserve the highlight detail.

[Ray]

Are you finding that you need to underexpose that much?  

Bobtrips,
Where I am, fairly close to the equator, sub-tropical at least, the sunlight is harsh and bright. Photographing pelicans for example, even when the sun is low in the sky, requires significant underexposure to avoid complete blow-out of the pelican's white feathers.

Each situation is different of course, but there's a general difference between 'far north' countries and 'equatorial' countries, as there is between winter and summer.

I remember vividly, when my now deceased father visited me in Darwin (from England) further north than where I am now, he was amazed at how bright the light was. He was 78 years old.

[Ray]

I think the issue of whether or not the ISO setting on a digital camera is accurate or not is quite trivial compared to the dynamic range issue. The fact that I can expose for the highlights with my D60, often producing an over all effect of significant underexposure, and yet bring out those dark areas, almost miraculously with Photoshop CS, without unacceptable levels of noise becoming obvious, is the critical thing for me.

Could you do that with the F828? I doubt it.

[Ray]

Matthew,
You're not being logical. I think we've now established that both the F828 and D60 (300D) are equally accurate, but in opposite directions. ISO 100 on film is equal to ISO 125 on the 300D and equal to (about) ISO 80 on the F828. Right?

A 1/3rd of a stop overexposure for the F828 might be necessary to get detail in the shadows, but there are going to be many situations where that results in blown highlights.

All the test shots that Ive seen of the F828, in bright conditions, have blown highlights. If it's a white brick wall, it might not be important because it's not obvious that there's invisible mortar but if it's white feathers on a bird, it's obvious.

Expose for the highlights with the F828, and I think you're in trouble.

[MatthewCromer]

Ray,

I shot a bunch of images of landscapes that included patches of snow and not one blew out the whites in the snow, unless I applied positive exposure compensation.

Now if you are talking about images in bright daylight, well then yes the 828 and all other digital cameras will blow out images under some conditions.  However, I find such images tend to bring very little to the table in terms of photographic quality to begin with.