HDR image to LDR using edge-preserving multiscale decompositions | Hacker News Reader

HDR image to LDR using edge-preserving multiscale decompositions

mathworks.com

46 pointsgprasanth27 comments

The source paper: http://www.cs.huji.ac.il/~danix/epd/

As with so many mediocre tone mapping functions, `tonemapfarbman` overemphasizes the red channel. I have seen better (without parameter tweaking).

The 2018 solution would be to just train a neural network to do the color mapping and call it a day without ever really trying to solve the problem.

Hell, it would probably work insanely well.

What would you train it on, without already having an algorithm you consider 'good'?

Pretty cool. I was a huge Mathworks fan for years, (still am, though I think Julia has a brighter future), and they always have great documentation.

The OS spotter in me is having a hard time identifying what's on the monitor in the image - is that some stripped-back Windows 7 verison or XFCE?

I'd say Win 7 without Aero.

Confession: I have yet to buy my first HDR capable camera...

If you can do multiple shots with different exposures (one after the other), you are HDR-capable. Software does the merging. Though some cameras do have automation support for that. The trees in the demo image look blurred because the branches moved between shots.

Ubuntu/Debian users, put this in ~/.local/share/nautilus/scripts to have it accessible from your right-click menu:

  #!/bin/bash
  align_image_stack -a BATCH -C $(echo -n "$@") && enfuse -o ~/Desktop/align-and-fuse-$(date '+%Y-%m-%d-%T').tif BATCH*.tif && rm BATCH*.tif

Prerequisites are align_image_stack and enfuse packages (Hugin users already have them or do it manually) and after that, just shoot bracketed and multi-select each set of bracketed photos > right-click > Scripts > run script. A timestamped finished image in TIFF format will output to your desktop. I usually open these in Darktable for minor post processing, then final export to JPEG.

Aren't really all cameras HDR capable with bracketing, i.e. shooting say 3 images or more with different settings.

They need some method for keeping the camera in the same spot, and (roughly) pointing in the same direction.

A smartphone without a dedicated funcion probably has too much delay between shots for HDR. Same for point-and-shoots without a tripod mount.

Even if you don't have/can't use a tripod , software stabilization is pretty good for handheld use. And there are a few ways you can pull that with OpenCV, so it's not that expensive editing software is involved (of course, Photoshop has it, and in a "photographer friendly way").

Stabilization deals with orientation. Location cant really be corrected for based on a 2D image.

Yes -- but it still works fine for the common distortions that a handheld camera introduces (assuming you don't go into very large time spans).

A 'poor persons' HDR can be done with ONE good photo. Load it into a decent photo editor, and adjust the exposure down two stops and save it, then repeat again but up 2 stops. Now with the 3 copies, run it through a HDR generator. Not as good as bracketing directly in the camera, but can yield acceptable results.

No. This will give the visual effect of HDR (which many think is something to be avoided - HDR should be a tool, not a style), but does not increase the actual dynamic range.

True, but based on the fact that hardly anyone has actual HDR capable displays, the net effect of the output is mostly the same.

It does.

My RAW Images has a higher dynamic range than the final image shows without any HDR feature.

You only put back what was already there and 'lost'.

That only works if you're shooting RAW. And even then all you've accomplished was a convoluted process to achieve the exact same thing you'd get in Lightroom by adjusting the Blacks and Whites sliders, and dragging the ranges horizontally in the histogram.

That does not invalidate my original statement.

I would argue that the end result should look slightly different.

But you do use more of your dynamic range

Really, if you are going down that road I prefer manually blending the different exposures in Photoshop. 5 years ago, that is how I dealt with bracketed exposures as well. Back then HDR still had a very definite look which I really tried to avoid.

I've combined hand held bracketed shots. There are problems at the edges of things due to slight shifts in location and parallax but this can be handled in photoshop and HDR (I used to use Photomatix) does a good job at removing ghosting and edge artifacts.

Was thinking the same thing.

Essentially the person is saying they don’t own an SLR (digital or film) or a mirrorless camera, don’t own a smart phone, and don’t have a pocket point&shoot made in the last ten years or more.

You need to be able to change only shutter speed or ISO, while holding aperture constant. That's not a problem for phones, because they're almost always fixed aperture anyway. And of course any DSLR or mirrorless camera can do it. But some point-and-shoot cameras won't have this capability, they'll want to be fully automatic, not giving the photographer that specific control.

The many point-and-shoot cameras are also fixed-aperture (they use an NDF in bright light rather than adjusting the aperture).

Why not a single exposure? My camera (Nikon D750) shows up as having 14 stops of DR on DxOMark and IRRC a typical monitor in ambient condition is 8-9 stops. I haven't tested any of this though. Any in-camera JPG processing is applying a tone-mapping function to compress the range. Is the HDR label reserved for an image that has more range than the human eye can see and not necessarily the viewing device?

Depends on the style of HDR; if you want that post-appocolyptic look vs a just a well exposed shadows and protected highlights look. Shooting ETTR will allow for the shadows to get extra exposure while using the highlight recoverability of a RAW file to protect the highlight. Each camera sensor is different, so you have to practice with your camera to see how much you can push the exposure before unrecoverable clipping occurs. Obviously, the camera’s preview will looked overexposed, so people you show it to will think you’re crazy.

Reply on news.ycombinator.com