HDR compression - Yuanzhen Li subband technique

Get ready to get your nerd on.

This is another of the many HDR compression algorithms that uses clever math to control compression and attempt to minimize artifacts, such as halos. Here is the developer page:

Compressing and Companding High Dynamic Range Images with Subband Architectures

Yuanzhen Li, Lavanya Sharan, & Edward H. Adelson

Dept. of Brain and Cognitive Sciences, and Computer Science and Artificial Intelligence Laboratory
Massachusetts Institute of Technology, Cambridge, MA

http://www.mit.edu/~yz​li/hdr_companding.htm

What is cool about this technique is that the developers have made the code freely available to any joe like me who wants to experiment with it. The code is written to run in a Matlab-like environment (you can use Matlab, or its open source, free cousin Octave).

The workflow I used is convoluted, but I warned you at the beginning of the thread....

I shot 9 to 11 exposures for each scene - I did this because it is easy to set up on my Promote Control and it was about 25°F out at the time and I did not feel like trying to edit the settings - in fact the LCD was slow to update, just to give you an idea of the shooting conditions.

I used three of the exposures and fed them to Zero Noise. From ZN i got a 16 bit per channel TIFF.

The compression code requires a portable bitmap/float map file format, so I took the TIFF into PS and converted it to a 32 bpc PBM.

Then you run the algorithm. Out pops a LDR image that has been compressed - the output from the default settings is a little too saturated. You can control the amount of desat by prescribing non-defaults, but I am not there yet.

So, back in PS I desaturated the image and did some final edits. Here are the results. I tried a few different kinds of scenes and found the halo artifacts to be minimal, although there are some areas where they are obvious. Again, once I try adjusting the parameters away from the defaults, maybe this will be able to be tuned for the image content.

/nerdage.

kirk

You a nerd?

I've got say though, your examples seem to be very, very realistic. Up until now, SNS-HDR has produced the most natural and accurate results I've seen. This is even better.

Personally I'm not a fan of bright sunlight hdr just does nothing for me. Is tha vally forge park? I went out last night and shot the art museum. Froze my ass off too /:

HastyPhoto wrote in post #13718314
Personally I'm not a fan of bright sunlight hdr just does nothing for me. Is tha vally forge park? I went out last night and shot the art museum. Froze my ass off too /:

Let's ask Kirk to show same size images from the best of the input shots and the output side-by-side.

I think this is amazingly good HDR work.

My only minor reservation seems to be that I think I'm seeing visible noise in the output (I see the same effect in my SNS-HDR outputs) I'm not sure if it's truly noise, or just a toning/contrast characteristic that makes it look like sensor noise, but sometimes I see a "graininess" to the output. Still, a 1000% better than the standard camera output....

The cabin sure looks like VF.

I like the results of this approach.

Thanks folks.

I will post comparisons as OldBaldy asked. In the image of the statue, I added a processing step that emulated (i can't remember which) film and blended that B&W film image back into the color image. That included film grain, and the grain is obvious throughout the image, not just in the shadow tones.

I should emphasize that these images are not the direct output of the compression function, but the compression that is applied provides a really nice starting point for processing in Photoshop. This is typically how I work, so i rarely see the toning or compression process as the final step in HDR processing.

It is Valley Forge. The first through third images are from Washington's Chapel - this location is my de facto HDR standard test area because of the outdoor gothic arched structure that always has exposure to that open field across the roadway. That field is brightly lit when the sun is out and provides background illumination that is bright, as well as a reflection on the flagstone floor that is bright - these areas metered at about 1/2000s and 1/4000s respectively. The roofed arches have dark brown wooden paneling that is always in shadow and provide a shadow tone that is off the charts most of the time. If I recall, this area metered at about 1s, maybe 2s. I shot all of these images at f/5.6 and ISO 100 or 200 (I will look at the EXIF). The sky was a completely cloudless pale blue to almost hazy white near the horizon. It was freezing-my-ass-off cold and windy.

Once I get the hang of shooting for this processing, I will be able to work much more quickly, as I will likely acquire a lot fewer images per scene, say three images at 3 EV apart or something similar.

OldBaldy - take at look at the DOF I got at f/5.6 - you don't need to shoot at f/13 or f/16 to get that depth, you just need a good long focus distance to achieve hyper focal DOF. Once you start closing down the aperture thinking you need to, to get the large DOF, you start getting diffraction and starburst patterns on your specular's and you get soft images.

Experiment with your optics and see if f/5.6 to f/11 isn't a better choice.

Kirk

Good post. I use the results as a layer, when I do try HDR.

Very good idea to add grain; nice impact.

BTW, have you tried focus stacking? If so, what do you make of it?

kirkt wrote in post #13725124
Thanks folks.

OldBaldy - take at look at the DOF I got at f/5.6 - you don't need to shoot at f/13 or f/16 to get that depth, you just need a good long focus distance to achieve hyper focal DOF. Once you start closing down the aperture thinking you need to, to get the large DOF, you start getting diffraction and starburst patterns on your specular's and you get soft images.

Experiment with your optics and see if f/5.6 to f/11 isn't a better choice.

Kirk

I will certainly do that, Kirk....and take some more experimental shots myself, with different aperture, for comparison. Part of my reasoning for the suggestion for smaller apertures may be because the 17-40L that I use, is very partial to stopping down on the 5Dc, with what seems to be negigable difraction issues - at least until f/16 or so.
The 5Dc also seems to be one of cameras least affected by difraction (or rather, the onset of difraction issues seems to be at a slightly smaller aperture, than what may be normal. I don't know the technical reason for that.)

I do agree that the largest aperture you can get away with, for the DOF you need, probably makes more sense, as most lenses seem to give their best IQ on the edges and corners of the frame, at around f/8 - f/11 from the many resolution charts I've looked at.

SHorton - yes, I have done focus stacking, and I like it. You can use it for many different things, but typically for extending DOF, especially with a macro. The Zeiss 50mm f/2 makro-planar is an incredible lens and the ability to shoot the shots like I posted here and then shoot macro close-ups (like abrasions on the D-ring of a seat belt system I inspected recently) all with one lens is really fantastic.

The D-ring I shot needed to by shot from an angle, with the surface I was concentrating on receding away from the camera. I shot 9 images with the plane of focus retreating away from the camera incrementally until I captured the entire length of the area I was interested in. Stacked them and BAM! full image all in focus.

FWIW, although the Zeiss is a manual focus lens (I use live view and a hoodman loupe to focus) autofocus lenses can been driven to focus stack with the current (beta, still?) firmware of the Promote Control for some of the cameras that it supports. See here:

http://www.hdrlabs.com …aBB.pl?num=1311​206654/0#0

OldBaldy - MTF charts are great, but try capturing some scenes you would normally shoot at f/13-f/16 with wider apertures and see if it works for you! Use a test chart of some sort, high contrast edges (i use a print on white paper, with black large text, and shoot from a reasonable distance (say 5 feet to 10 feet) and roll through the apertures. Shoot so that the target is on a horizontal surface (a table) and the camera is positioned above that surface tilted downward at like a 30 -45 degree tilt below horizontal. You could shoot a yardstick placed parallel to the camera axis, so that the yardstick recedes away from the camera. Focus on the center of the yardstick (the 18" mark) and you will get DOF blur in front of and behind the focus point. Exposure and shoot, stopping down the aperture and using longer shutter speed to maintain exposure. Shoot under the same lighting conditions for all shots.

Then open the images and see how the edge rendering, the field rendering, etc work out for that lens/camera combination. See if your image that is in focus starts to degrade in contrast and acutance at some consistent aperture. THen take some shots out in the field with longer focus distances and see how much you can open the aperture and get the DOF you want.

Attached is an example of this target and position, using the 50mmMP, in this image it was set at f/4.

kirk

HOSTED PHOTO
please log in to view hosted photos in full size.

kirkt wrote in post #13727342
OldBaldy - MTF charts are great, but try capturing some scenes you would normally shoot at f/13-f/16 with wider apertures and see if it works for you! Use a test chart of some sort, high contrast edges (i use a print on white paper, with black large text, and shoot from a reasonable distance (say 5 feet to 10 feet) and roll through the apertures. Shoot so that the target is on a horizontal surface (a table) and the camera is positioned above that surface tilted downward at like a 30 -45 degree tilt below horizontal. You could shoot a yardstick placed parallel to the camera axis, so that the yardstick recedes away from the camera. Focus on the center of the yardstick (the 18" mark) and you will get DOF blur in front of and behind the focus point. Exposure and shoot, stopping down the aperture and using longer shutter speed to maintain exposure. Shoot under the same lighting conditions for all shots.

kirk

I did exactly this (with my steel yardstick) on the kitchen center island back in about 2009 when I got my first DSLR (20D). The DOF variances were astounding to me, using my Nifty Fifty...from 1/2" DOF at f/1.8, to the entire ruler in focus at f/22 or whatever the minimum focus was.

I use the A-DEP button all the time, to gauge DOF...find it very useful. That generally gives me a fair indication of maximum aperture for my subjects/scene/composi​tion/lens/camera combo.