"Patri’s First Law Of Statistics:
Correlation implies causation when convenient; otherwise, there are confounding factors."
- Patri Friedman
More pages: 1 2 3
HDR photography
Sunday, September 20, 2009 | Permalink

Back in February I made a blog post about HDR cameras. Since then a few things have happened. Pentax released their K-7, which is the first SLR with built-in HDR. It's taking the traditional approach of multiple exposure and consequently needs a lot of processing after shooting (12 seconds according to some sites). Recently Sony released the a500/550, which is both faster (2 seconds) and unlike K-7 has built-in image alignment which should allow for handheld HDR photography. Fujifilm also released the F70EXR, which is a cheaper variant of the F200EXR. Fujifilm is still the only vendor to use an actual HDR sensor and thus need no particular processing at all. All manufacturers are still only producing a final jpeg image unfortunately. Fujifilm would be in a position to offer a RAW format with HDR since they have it built into the sensor, whereas the other manufacturers could opt to use the EXR format. Anything that allows you to post-process the tonemapping would be great. But this is still early and I'm sure better solutions are around the corner and this is all a very exciting development.

After having been tempted for a while I finally pushed the order button and got myself an F200EXR. After having played with it for a while and learned how to best take advantage of the HDR feature I'm really loving it. And here's why:



The picture on the right has HDR enabled. The picture on the left is about what every other camera would produce. The pictures are taken at the same time with the same settings and no other post-processing was done other than resizing for the web.

I'm convinced that HDR will be the next big thing in photography.

Name

Comment

Enter the code below



Nick
Sunday, September 20, 2009

I do HDR the "hard" way, but am intrigued by a camera with a better sensor! There's an in depth technical review here -

http://www.dpreview.com/reviews/FujifilmF200EXR/

I use things like Photomatix, Enfuse, and Hugin to merge multiple exposures, and wouldn't mind being able to skip that step!

http://meshula.net/photos/2009SummerTrip/index.html

Overlord
Monday, September 21, 2009

I would like to see how good Single exposure HDR could get on a high end DSLR, it would make taking HDR timelapse a bit easier, and not to mention remove the colored fringes you might get on things that move in the wind.

Sean Barrett
Monday, September 21, 2009

There are two important aspects to HDR - more precision and tone-mapping.

The dynamic range of gamma-corrected 8-bit is already pretty wide, just not very precise. It would be interesting to see what one would get with the regular camera on the left with a faster shutter time (to capture the dynamic range) and then tone-mapped to match the right. Presumably it would have hideous banding, but who knows.

Mathias
Monday, September 21, 2009

@Sean If you would reduce the shutter time, there would be not much detail in the walls on the left and right. There is less detail on the HDR picture, that should be due to a sub optimal tone mapping. So all I can hope for is canon releases a camera with a hdr sensor, so with CHDK I can save raw images with HDR.

Wes
Tuesday, September 22, 2009

You might be interested in this work:

http://graphics.stanford.edu/projects/camera-2.0/

Where the group is "building an open-source camera platform that runs Linux, is fully programmable (including its digital signal processor) and connected to the Internet, and accommodates SLR lenses and SLR-quality sensors".

The sensor is from the N95 cell phone, so I would guess its not an HDR sensor. Still, its pretty cool work.

Overlord
Tuesday, September 22, 2009

@Sean "There are two important aspects to HDR - more precision and tone-mapping"

Well not precisely, Dynamic range is defined as the range between the darkest non black color and the brightest non white color detectable, on my D200 it's about 9 stops, increasing precision does increase the usable dynamic range slightly, especially in the dark areas, but it's still in the same general range.

and tone mapping is the process of transforming light in an HDR image to fit a LDR image, it's has nothing else to do with HDR, in fact you could tonemap a HDR to HDR and LDR to LDR.
But you don't have to.

So basically HDR is when you get an dynamic range much closer to that of our eyes, about 15-20 stops or so would do.

So i wouldn't say that the F200EXR is especially that much HDR, but it is a step in the right direction, i mean i could probably do most of the compensation in the above picture if i shoot raw and then fiddle with it a bit, but that takes a lot of work.

And about 8bit, well it's ok for normal monitors since anything more gradual than that is not possible to detect with our eyes, but that is assuming a maximum dynamic range of about 1:500(9 stops) like all lcd monitors have, if you go way beyond that, you will need HDR.

Sean Barrett
Tuesday, September 22, 2009

Overlord: gamma-corrected 8-bit at a 2.2 gamma has a range of values from 0^2.2 to 255^2.2. Ignoring the zero for the dynamic range calculation, the effective range is 1^2.2 to 255^2.2, which is a dynamic range of about 200,000:1.

I don't know a 'stop' is a measure of, but 200,000:1 is obviously between 17 and 18 powers of two, and it's well over the range identified by Debevec in his original HDR paper of the things he considered HDR -- in other words, existing 8-bit imagery met his definition of HDR.

The actual problem is that there just isn't enough precision in the darks; if you tone-mapped them brighter you'd see all sorts of crap.

Which is my point: I'm not saying 8-bit imagery is good enough. I'm saying if you're going to make these side-by-side comparisons, there's other interesting comparisons to make instead of just showing a blown-out 8-bit image.

Humus
Wednesday, September 23, 2009

Overlord, actually, the eye can detect smaller differences than 8bit. For a colored gradient it may be hard to detect, even if it's possible, but for a grayscale 8bit gradient the banding is pretty obvious. If you tried the 10bit sample in the August 2009 DirectX SDK you see a very clear improvement going from 8bit to 10bit, assuming your monitor does not truncate the 10bit to 8bits. Apparently my HP LP3065 is able to use all 10bits and the difference was plain as a day and I had to screenshot it to verify that there really was only a difference of 1 between the bands in the 8bit image.

Sean Barrett,
a "stop" in photography is a factor of two. So 200,000:1 would be 17.6 stops. However, the power of 2.2 is really just an approximation of sRGB. The real formula uses a linear part at the bottom and an offsetted power of 2.4 curve for the rest. So for the lowest non-zero value you have (1/255)/12.92, which then gives the dynamic range as about 3300:1, or about 11.7 stops.

More pages: 1 2 3