Shooting high ISO vs underexposing and lifting in post question

Started 3 months ago | Discussions thread
gardenersassistant Veteran Member • Posts: 8,917
Re: Shooting high ISO vs underexposing and lifting in post question

bobn2 wrote:

gardenersassistant wrote:

bobn2 wrote:

gardenersassistant wrote:

Slaginfected wrote:

lattesweden wrote:

Now the thing I don't understand is why does a camera manufacturer like Sony who have a sensor that is near ISO invariant use the first case at all and by that reducing dynamic range by pushing up the recorded signal? If a proper metering says the scene needs ISO6400 why not always record that image at ISO800 and just put in a metadata value that it was recorded with 3 stops below what the metering said to tell all RAW-converters (including the built in one) to lift the brightness 3 stops when producing the JPEG-image (or whatever format is choosen) and by that have 3 more stops of DR available?

It is very telling that all the people here are completely incapable of dealing with that question. That is one of the reasons why I have to butt my head time and again against a wall of concrete, and over the years it finally starts to show cracks! (And I will continue, until I made it through that wall of concrete.)

The core issue here is that for most things stop with the RAW data and then there is some "magic" and final image. Which results in things like that there is no difference between ISO800 pushed up 3 stops and ISO6400, if we ignore the additional headroom available in the ISO800 image.

Sadly, that is not correct. I could start with discrete math with set theory and all that garbage following, but it won't lead anywhere. Things can be explained in a much simpler way. Even though it may not be fully correct compared to the math variant, it is a fair enough abstraction of what is going on and way easier to understand.

Imagine you have an image recorded with ISO800 in 14 bits. The data shows maybe like this, where x denotes a "used" bit:

[x|x|x|x|x|x|x|x|x|x|x|x|x|x]

Now, if you push it up 3 stops, highly simplified by shifting the bits, the following happens:

x|x|x [x|x|x|x|x|x|x|x|x|x|x| | | ]

As you can see, the headroom moved out of the 14 bits, can be considered "overexposed", and at the low end there is no information, because we didn't record it. Now lets compare that to an ISO6400 recording:

[z|z|z|z|z|z|z|z|z|z|z|z|z|z]

It doesn't have the additional headroom, but! it has information in the lower bits.

The results of that can actually be seen if you overdo the pushing; with some experience and where to look 4 stops with 14 bits can be more than sufficient enough already. The resulting artifact is called posterization, as limited amount of tonal steps. A lot assume "banding" with this term, but not all posterization shows as banding. If you know about oldschool computer stuff with their limited amount of colors and dithering, you may get an idea how this will look like.

That appears to be consistent with what I found when I experimented with this. As I wrote here (see the emboldened section at the end below - the preceding text is context):

gardenersassistant wrote:

It is looking like this ISO 100 experiment is drawing to a close. After four test sessions of around 1500 captures in total I have come to the conclusion that my initial impression of advantages with highlights were in fact mistaken. Despite what I have read about lowering ISO (with a fixed exposure) to protect highlights I have not seen evidence of this effect with my images.

I'm not arguing that the effect does not exist, simply that I have not seen evidence of it with my own images. It may for example be [hunch] because I am using illumination which is low (in the context of the small effective apertures I am using) and I may not be reaching the point of [terminology?] overloading any of the pixels. On the other hand, there seem to be very visible problems in shadowed areas; not just dark backgrounds (where I'm seeing posterisation after shadow raising), but undersides/shaded areas of insects where shadow raising is producing very crude results (for example tending to undifferentiated black) which are distinctly ugly to my eye, and which I have failed to deuglify with modified (global or local) processing.

Your results don't coincide with the results obtained by many others, and the test protocols on this site. The fact that you see posterisation would indicate that you are using a bad technique for 'shadow raising' most likely doing the operations in post-processing, when the information from the lower bits of the raw data has already been discarded.

Could be. PhotoLab does the raw conversion and also does shadow raising. How its internal pipeline works I have no idea. I sometimes do additional shadow raising in Lightroom, on TIFF files coming from PhotoLab.

I just went through a couple of the experimental sets in Fast Raw Viewer with Exposure Correction turned on. This increased the lightness by around 4 stops on some that I looked at and I didn't notice any problems. So the issue might well have been somewhere in my processing pipeline.

That's one of the reasons that I like to distinguish between 'processing' and 'post-processing'. Even if the processing is to a 16-bit TIFF whatever is below what the processing options determine to be 'black' will be chopped off. The problem with these modern powerful tools is that the mix up the two stages, and it's very hard for users to know what's what. Very often the exact result depends on which of the very many different tools and options you use.

If I'm trying to do this kind of thing I prefer to use very simple raw converters which are really only trying to be raw converters. Sometimes I'll produce a TIFF configured for further editing, which means getting the white balance right,

Yes. For these I give the processor (PhotoLab) a white balance to use taken from a calibration shot for the diffuser setup I'm using. This works nicely as flash is the dominant light source.

but setting the black point very low (like zero) so that no information gets cut off.

Interesting. Is that doable with a Blacks slider (presumably pushing it all the way to the right, and the images look very flat), or does it have to be strictly Black Point in Levels? (Don't think PhotoLab has Levels. Anyway, I feel an experiment coming on. :)).

They look odd, but then you can take them into your favourite picture editor and us the interactive tools to get them as you want.

The problem with ISOless workflow is that is it is sufficiently different from the workflow that commercial tools are designed to support that they usually do a bad job of it.

I can well believe that might be the cause of some of the results I was getting with four stop and more lifts.

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow