HDR processing option for PWP?

[MarkT]

I know nothing about how to implement something like this, but one of my students used ChatGPT with a series of images to produce an exposure-blended result. I tried this myself and see that ChatGPT used something called "average exposure fusion" to process the images.

A quick search turns up this site: https://learnopencv.com/exposure-fusion ... pp-python/. And also this paper: https://web.stanford.edu/class/cs231m/p ... fusion.pdf

So, I'm curious if PWP could integrate something similar?

[pierrelabreche]

Expand dynamic range transformation

[MarkT]

I was thinking more about an augmented form of Stack Images.. load a series of images and then run exposure fusion on them.

[pierrelabreche]

Expand Dynamic Range
As good as modern camera sensors are, there is a limit to how much dynamic range they can capture. When
photographing scenes that have a large range such as landscapes with a bright sky and dark foreground, you
may need to choose whether to expose for the sky and get a very dark foreground or expose for the
foreground and get a washed-out sky.
This transformation lets you create an image with an extended dynamic range by combining a normal
exposure with either an overexposed or underexposed version of the same image, or both. By using an
underexposed version, you can recover clipped highlight detail in the normal exposure, while an overexposed
version lets you recover clipped shadow detail. This only works if all the images are perfectly aligned (e.g. shot
using a tripod), the subject does not move, and the images are processed identically. For best results, try to
use this transformation early in your processing workflow.
As a rule, an extended dynamic range image has less contrast than any single exposure, similar the way color
negative film has wider latitude but less contrast than slide film. This transformation lets you apply a final
shaping curve to the output image which you can use to adjust its tonal balance if you wish.
The normal, underexposed and overexposed images must all be the same type, have the same size and have
the same profile. If both the underexposed and overexposed images are omitted, the result is simply a copy of
the input image. You may specify either an underexposed image, an overexposed image or both.
A similar effect can also be created using the Stack Images transformation, but Expand Dynamic Range is
easier to use and does not require calibration images.
This transformation supports two methods for merging bracketed exposures: HDR and Exposure Fusion.
They usually produce similar results, but depending on the input images, one method may produce more
pleasing results than the other:
HDR
This method works by first compressing the dynamic range of the normal exposure to make room for extending
the final dynamic range. It then attempts to map all the images to this expanded range, using the under- and
over-exposed images to fill in the gaps at the bottom and top of the range. When you use this method, you can
select how much to compress the normal image by setting its white and black points. The more you compress
the normal image, the lower the resulting image contrast, but the wider its dynamic range. To accommodate
the wider dynamic range, the output image is always created with 16 bits per channel, even if the input images
have 8 only bits per channel.
Exposure Fusion
This method blends the normal, and over- and under-exposed images, attempting to combine the best parts of
each one. It uses a combination of three adjustable criteria to determine which pixels to emphasize and which
to de-emphasize:

[MarkT]

Right, thanks Pierre. I had forgotten about Expand Dynamic Range. Although I remember trying it in 2023, I found the results weren't what I was expecting so I haven't used it since. I tried it again today, and remembered why I wasn't fond of it - the limitation of using only 3 images. Many scenes that can benefit from HDR processing usually have highlight-shadow values ranging upwards of 15 or more stops of exposure difference, and that is difficult to capture in only 3 images. I regularly use 12 to 16 images to capture HDR scenes (and process them with SNS-HDR Lite), and since the Stack transformation can handle 15 images my mind went to adding 'exposure fusion' to the list of blending options in that transformation. Also, as I understand it, there is the inclusion of minor alignment functions available, which Stack and EDR do not currently support.

Anyway, thanks for reminding me about the EDR transformation.

[pierrelabreche]

Hi Mark,

HDR stacking could be more usable if it accepted a flexible number of images to be stacked ( 1, 2, 3, 4, .... ) and with auto alignment of the image set.

I rarely do HDR stacking, I have used occasionally Nik HDR Efex and Luminar Neo.