Averaging option for white balance picker

As Alexander points out, the bigger the better, on a solid color. That is my experience too, especially if compared to applications which allow a variable sample size.

I pick white balance on white or grey shirts/fabrics, rocks, clouds more often than on a grey card, simply because the type of photography I do, and the results of the C1 picker are rless than ideal (v15.2), not consistent, gives me bluish or yellowisch results quite often and almost every virtually click results in a white balance differing too much from the previous click.

On contrary, the white balance (and color picker) in darktable provides me very consistent(and good) results, it is a rectangle you can draw as big as you like, even full image size, which then works as the auto white balance of a camera, averaging the whole scene.

Bottom line:

If you ask me, and the other users in this thread, C1 is not the gold standard for this matter. For all but grey cards, 5x5 is not enough, but bigger size requires targets which are big enough in the image, so a variable or configurable size would probably be best. 

Hi Alexander,

Thanks, that matches the information that MS Copilot has been suggesting this morning as I researched opinions.

There seems to be agreement that there is some special benefit where noise is concerned. However, so far I have not discovered anything, that seems to be directly related to the subject, suggesting that larger than 5x5 might be even more useful.

I'll come up with some more questions to pose to AI trying some slightly different phrasing to see what results are presented!

BeO,

To answer your question, it may depend on the sensor, the nature of the image (amount of "noise" for example) and other factors.

For a specific sensor array type, in theory, if you can find the "perfect" area of the image to assess where, for example, 18% grey exists, then a small sample in a busy, highly variable image, may be the most accurate way to find precisely the right value. 

Make it bigger and noise factors might creep in.

On the other hand with a very noisy raw file processing with a larger array my permit the calculation to discard noise.

And presumably different sensor types may require different selection strategies, in an "ideal" situation.

The challenge is to find an ear that is representative of the entire image in terms of a genuine "white" value (or a proxy for that value) that is not just a colour cast influence with limited effect (no matter how apparently obvious the colour effect may appear to be.)

The WB adjustment should be a specific way of setting an acceptable baseline value for White from which any desired colour and tint adjustments can be made using appropriate tools.  Basically WB is a means of allowing for the sensor's differing sensitivities for different colours in different situations. 

368752449678 it was to ensure consistent results when using the picker on a color target, even when some sensor noise is present. 3x3 was too inconsistent, and 5x5 improved the results quite dramatically. When using the picker on a solid color, a larger sample size will always be more accurate than a smaller one.

Hi Alexander,

Was the motivation to move to 5x5 partly driven by the additional processing power available in modern computers or, perhaps, the greater number and capability of pixels available in recent, larger processors?

I'm mindful that the idea of assessing, or in the case of this discussion, re-assessing WB during editing, is likely a more complex subject overall for developers and system designers than might be evident at the UI. It is quite easy to get sucked into using WB as a one-stop metaphor for fixing a colour cast to a personal preference but it may often be difficult to constrain the concept to finding a good WB base point from which to start the colour cast elimination as a separate adjustment using different tools.

OK, thanks.

Ian

368074128197 it was changed in 13.0.0 from 3x3.

Thanks, 374153015698.

When you say "as it is today" does that mean that it's changed from what it used to be, or has it always been 5x5?

Ian

Thanks for the post!

As it is today, the WB picker samples 5x5 pixels, and not just a single pixel like some of the comments indicate.

I think the word "option" here is the key. If you want to pick WB from a single pixel, you could. But in my experience, an apparently neutral area on an image from which you might pick a WB are not always as uniform as they first appear, and the WB you get depends very much on just which pixel the pointer happens to hit on. (Especially so in a noisy image.) So if that difficulty might be ameliorated by an option to pick the average of a small number of pixels, why not? 

(The same goes for other colour picking tools such as the colour editor too.)

Ian

Surely a larger sample would tend towards LESS accurate WHITE balance rather than more accurate?

Why do you think so?

Surely a larger sample would tend towards LESS accurate WHITE balance rather than more accurate? NOt so useful for people doing, say, product shoots and trying to get entirely accurate colour matches.

However, it might suit anyone trying to reset an assessed white balance towards personal preferences, though whether that is the most effective way to make such an assessment may be the subject of great debate.

I guess the White Balance tool is sampling the Raw data?

That's what I would expect.

Can you shoot the grey card/color checker out of focus then refocus for your final capture? I often defocus to shoot scene references if the target I am shooting has any texture. Colorama unrolled in front of a painting etc if on location.

What about travel photography or other genres where you neither have grey card in your image nor a series of images, not even to speak of blurry images...

Can you shoot the grey card/color checker out of focus then refocus for your final capture? I often defocus to shoot scene references if the target I am shooting has any texture. Colorama unrolled in front of a painting etc if on location.

The ability to choose a larger sample with the pickers would be great. Some kind of averaging or smoothing of the scene reference shot before the uniform light calculation would be good too. Users would normally want a gradual correction with this function rather than a pixel by pixel correction (unless using for sensor spot removal)

*Note. I just experimented with 'blurring' the image using negative values for Clarity-Structure which does not affect brightness or contrast. Also setting dehaze to -100 and sampling the square on the Colorchecker Classic I am using with the Haze shadow tone picker so the brightness of that tone isn't changed. This is as blurry I can get the image without changing the brightness of my sample (patch 20 in my case) but when sampling across the patch, the changes in White Balance are as different as with no digital blur added. I guess the White Balance tool is sampling the Raw data?

Yes, please improve the picker as sugested, with some flexibility to the user, i.e. a  configuration or drawing a rectangle or circle. In its current state it is rarely fit for the purpose, only under ideal lighting and noise conditions and with a grey card or similar.

I agree - this needs to be addressed. The White Balance picker is frustrating because using it can result in surprisingly large swings in tint and temperature from one pixel to the next due to sensor noise and other unimportant variations. The user needs to be able to select a sample size, perhaps by clicking and dragging the picker tool to create a sampled area, or just set a reasonable default like 3x3 or 5x5. I could see both mean or median sampled value from an area working well, but single pixel sampling is almost never what the user wants or needs.