What I learned from Gollywop -- and what I wonder

Started Mar 26, 2013 | Discussions thread
Anders W
Forum ProPosts: 18,408Gear list
Like?
Re: What I learned from Gollywop -- and what I wonder
In reply to Macx, Mar 29, 2013

Macx wrote:

Anders W wrote:

Macx wrote:

Anders W wrote:

Macx wrote:

olliess wrote:

Macx wrote:

Even if we presume the same amount of exposure, the lower averaged noise would still give an advantage to multiple captures than the single one.

Well maybe here is a way to think of it:

Your goal is to capture an image without clipping, but with a certain shadow area "out of the noise."

The "ideal" exposure would simply be a single exposure which gives enough exposure to the shadow area that you want to capture. For our purposes, let's assume we can't do this because the sensor has an insufficient DR.

We could stack N unclipped (ETTR) images, which is certainly advantageous compared to a single unclipped exposure, but we also accumulate read noise as sqrt(N), so it is not perfectly efficient compared to the ideal exposure, where the read noise counts only once and we are only collecting photon noise with increasing exposure.

No, I agree it isn't perfectly efficient, but I think there are a couple of places where it could shine.

Firstly, if we want to increase highlight details, I don't see any way around it. Admittedly, I can't think of many times where lack of highlight detail is a problem, but even so.

While stacking without bracketing does increase highlight detail while the bracketing approach does not, I can't think of single example where there is much to be gained from reduced highlight noise. From a noise point of view, this is the least problematic region of an image, the one where SNR is highest.

Indeed. But this isn't just about the extreme highlights, though. If the bracketed shot is -3 EV, that is 3 EV of the dynamic range with blown details.

Not sure I follow you here. What do you mean when you say that the bracketed shot is -3 EV? If you bracket, you would start with an ETTR exposure and then increase exposure from there, usually by one EV (or less) at a time. No part of the dynamic range will have blown details as a result of that. The topmost EV of the scene would come from the first image, the second topmost from the first and second image (with the second weighing in more heavily) and so on.

No, you're right, I got that mixed up. I was thinking that the second shot was metered for 3 LV's lower and as such it should of course be shot at +3 EV. Nevermind. Regardless, your method is of course a better practice than just using such a two-shot exposure, but I don't think it's that uncommon that people pick as few captures as they believe they can get away with.

Btw, I'm not being unappreciative about the benefits of bracketing, I originally underestimated the amounts of exposures needed to get an EV more of DR, and standing corrected, I'm now just trying to explore what advantages it might bring to use stacking, and break my ...what's the word for vanetænkning? ...business-as-usual approach, and try to think a bit out of the box.

And what I was thinking was that stacking might be a way of improving noise performance beyond the maximum exposure at base iso. But I'm not sure I understand the fundamentals of merging bracketed shots, because I assumed that you just picked the best of the captures for that particular part of the tone ...map? But if all the data from the not-best captures for that particular part of the tone map is then discarded, that is of course a waste... so when we merge bracketed shots are we in fact stacking the captures also, just shifting them so they're added in the right place?

This depends on the software at our disposal and what we tell that software to do. What LR/Enfuse would normally do is to use all the shots that are at all useable for a certain area of the frame but weigh the shots differently depending on how optimal their exposure for that area is. This is the best approach if the scene is perfectly static. But if it is not, it may be preferable to tell the software to use only one shot (the best) for each part of the frame.

Consider for example a landscape where the foliage is moving slightly. If we are lucky, the movement is slight enough that it would be frozen by the shutter speed required to expose that particular area of the frame optimally. However, the foliage would still move sufficiently to look somewhat different in one frame than another. If we used all the frames to render the foliage, it would be blurred (in a chopped-up, ghosting kind of way). But if we use only a single one (the best for that particular area) it will look fine.

Note furthermore that while continuous weighting of the shots is better when the subject is perfectly static, it is not vastly better. The saying that no chain is better than its weakest link is very relevant when it comes to images. Any ordinary image has higher IQ in the highlights than in the shadows. A major advantage of bracketing is that it allows us to quickly improve the quality where it is weakest (in the shadows). And this advantage remains regardless of whether we use continuous weighting of the shots or a single shot per area of the final image.

Stacking without bracketing, by contrast, adds a lot of good data per new shot in areas of the frame that are already excellent (the highlights) but rather little in those parts of the frame where it is poor (the shadows). Furthermore, it is of no help at all for scenes with slight subject movement, like the landscape with moving foliage that I described above. If we can't use more than a single shot for each area of the frame and the exposures we can choose from are all alike, we are effectively back to square one: a single shot.

Nevertheless, there are cases where stacking without bracketing is preferable, such as when we can't use a tripod and the shutter speeds required for proper bracketing would yield blur due to camera shake. Another case where stacking without bracketing is useful is when you have objects in the frame that you'd rather get rid of and these objects are moving (e.g., cars or people). If they are moving quickly enough, or if you wait a bit between shots, these objects will effectively disappear in the merged image, particularly if you take a pretty long series of shots.

Eh, I apologise, English isn't my first language, and my limited lexicon is failing me. I'll try to explain it with an example: We have a scene with a dynamic range of 3. In order to get a better image we bracket by making three captures with a shutter time of respectively 1, 2 and 4 seconds.

Question is then, when we merge the captures, are the shadows going to use only the 4 second shutter time, or are we going to use the other captures too to get the maximum exposure for this bit?

If we are, there is little extra advantage to stacking, because we're already combining it with the bracketing. If not, it seems to me, we should.

Secondly, it seems that for the E-M5, stacking for example two ISO 400 captures gives less noise than one ISO 200 shot. The thing to note about this, is that this is with the same total exposure time. Here we're not talking about taking the ideal exposure and then another to bracket or stack. Here we're replacing the ideal exposure with two shorter exposures that yields a better result when stacked than the ideal one.

This is a rather special case, though, that rests entirely on the fact that read noise is nearly cut in half by going from ISO 200 to ISO 400 on the E-M5, making the DR from a single shot at ISO 400 very nearly the same as that of an ISO 200 shot at double the exposure.

Even so.

Even so what? What does this demonstrate?

Well it demonstrates a case for using multiple exposures instead of a single. I'm not sure I understand what you mean? Even if this case applies only to the E-M5, it's still remarkable, isn't it?

Well, if you find the drop in read-noise between ISO 200 and ISO 400 remarkable, I guess it is.

Thirdly, and this is a subjective thing, but I wonder if it's easier to get a "realistic" look from this way of doing HDR than the normal bracketing practice that sometimes produce a bit surreal images.

Whether the image looks natural or not has little to do with whether you bracket or not. It has to do with the way the images are merged and tonemapped. There is nothing unnatural about the bracketed example I already linked you to in my previous reply here:

http://www.dpreview.com/forums/post/51173651

To a large extent that is true, but this is why I wrote "easier", since a stacked exposure is, practically speaking, a single exposure and as such needs no special considerations when merging or tonemapping.

So what is more difficult or requires special consideration in the example I linked to?

I wouldn't know. I just assumed, wrongly perhaps, that there was some sort of extra user input required for merging files into a "hdr"-file, where none would be needed for stacking the same exposure.

Not really if you use software like LR/Enfuse. If you use software where you first create a 32-bit HDR file and then do the tonemapping separately, it may take more effort but give you more/other options as well. If all you want is a natural-looking, low-noise rendering with minimal work on your own part, I would think that the Enfuse approach is the way to go.

Now suppose the scene can be spanned by exactly two bracketed exposures, which we combine. The first exposure is the same ETTR as above, and then we expose a second time with the "ideal" exposure which clips the highlights but captures the shadow area with an adequate SNR. The first exposure incurs "1" fixed time cost, while the second exposure improves as the "ideal" exposure, and therefore this method may be more efficient compared to stacked exposures.

It's probably more practical in most situations. I think my personal yardstick is going to be the summed total of the length of the exposures when considering the advantage to multiple captures over a single capture.

Sorry, but why would that be a good yardstick? For non-bracketed stacking, the improvement in DR (as measured in EV) is ordinarily proportional to the square root of the total exposure time, or equivalently, the square root of the number of shots, so that four shots gives one EV more, 16, two EV more, 64 three EV more, and so on. For non-bracketed shots it gets a bit complicated because it depends partly on the number of shots, partly on the exposure difference between these shots. As a very rough rule of thumb, however, I would say that if the exposure difference between each shot is one EV, you gain one EV worth of extra DR per added shot.

It's a good yardstick because it tells me if it's efficient use of my time or not. If I can get the same result by using two methods, but one method takes considerably longer, I'd take the less time-consuming one. If on the other hand, two methods takes the same time, but one method yields a better result, I'd pick that one.

So what would you say takes longer, all things, including PP time, considered: five shots with bracketed exposure at 1/400, 1/200, 1/100, 1/50, and 1/25 or 31 shots at 1/400. Total exposure time is the same.

Keeping count of 31 shots might be stressful. I wouldn't think post-processing should take very long in either case? Computers are usually good at that numbers business, and if anything it could be done in the background.

Even if you just count the user-input time at the PP stage, not the time it takes for the computer to do the number-crunching, processing 31 shots takes significantly more time than processing five. And the fives shots would give better results.

I also find it useful, because I think it's a fair assumption that we're always somewhat limited to the maximum total shutter time that is either practical or desirable.

We are limited by time all right, but shutter time has a rather weak relationship with total processing time (when shooting and afterwards).

Oh, definitely. But the lab work isn't restricted by subject movement, changing light or limited access time. Even if we assume the stacking or merging bit of the post-processing takes hours, it can be done while we sleep for example.

I'm curious what else you would suggest to guide the choice between taking one, stacking or bracketing? I guess you could assume we had unlimited time, but is that ever really the case?

I would weigh processing time (rather than shutter time) against quality improvement (and the need/desirability of such improvement).

It should play in, I agree, but in my opinion it isn't as important as how much time we use in the field.

Another way of thinking about it in the case of bracketing with a separation of one EV between shots is that the number of EVs with a quality just as good as the brightest highlights in a single capture is equal to the number of shots. So if, for example, you make 10 exposures and increase exposure by one EV at a time, you can render the ten topmost EVs of the scene with the same quality that the highlights have in an ordinary shot. Then, with the E-M5 at base ISO, you would have 11 EV worth of DR of ordinary quality below that point.

 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
Reply   Reply with quote   Complain
Post (hide subjects)Posted by
BothNew
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow