odd effect in Levels and Color

[MarkT]

Sounds like you have the right idea Maciej. I cropped out a small section (reducing the total number of pixels) and the effect goes away (or at least is reduced so I don't notice it...).

I've sent a couple of images to Jonathan.

Thanks for the support.

[jsachs]

I checked it out and when reducing the Levels and Color dynamic range threshold by a factor of 10 the bright spot was detected correctly and it worked fine. I think the current threshold value works well for most images so I am reluctant to change it. Given that Levels and Color is a simplified transformation I would also prefer not to add the ability to adjust the thresholds.

As I mentioned earlier, Levels and Color uses a couple of thresholds to make guesses about the input image -- one for dynamic range and a similar one for detecting the auto white balance. These guess can fail in extreme cases and produce unexpected results. The Adjust transformation makes no guesses and can be used in these cases.

[tomczak]

I agree with not changing anything in L&C, as it works well and it's simple and there are just extreme cases that make it miss-guess the white point (as in astrophotography). I have a suggestion though, but not sure how good: instead of using the average of the 0.01% of the brightest pixels, wouldn't it be kinder to extreme cases like Mark's to use the median (or even Tukey's tri-mean, although I'm not sure if it would be better then median) of them?

[jsachs]

Using the median does not work very well -- for example, if there is a large red barn or a lot of blue sky in the image it will try to remove these colors.