Sunday, April 18, 2010

RAW Conversion: what's in a histogram?

I shoot Canon RAW exclusively. Last time I directly shot any jpegs it was by mistake, and it was a total disaster -- or almost.

But this means that the RAW files need to be converted into something that's human-decypherable: RAW files in-and-of themselves are inherently non-viewable without some sort of intermediate translation applied.

For RAW conversion I use Breeze System's BreezeBrowser Pro: I'm not professional enough to be able to afford the Adobe Tax(tm).

The single element I work with most frequently (which is to say, on every single photo) during RAW conversion is Exposure Compensation -- and this is adjusted by refering to the histogram for each photo before I convert it.

Here's three micro-adjustments made to one photo before conversion (clicky):

EC_comparison
The range of Exposure Compensation adjustments ranges from -1.3 Ev to -1.5 Ev in .1 Ev steps.

Notice that the histogram is greatly similar across all three adjustments, but with a slight and significant difference.

Essentially the histogram represents a graph depicting 255 steps across the range of values within the photo from 0 == black to 254 == white. The taller the vertical bar for a particular value, the more pixels within the image at that value point. Bars at the two ends represent data that is out-of-range and theoretically lost to the image.

(I say theoretically because it is possible to "pull down" the bars at either extremes and redistribute data back into the low or high end of the histogram using the slider.

I'm not sure if the formerly "lost" data in the RAW image is actually recovered or if just an illusion of data recovery takes place, but as the end bars are pulled down the histogram across the usable range plumps-up, so I'm happy).

Most theoretical discussions of histograms talk about lovely symmetrical bell-curves shaped along a standard normal distribution.

That's lovely and all, but I almost never see anything like that in reality.

In fact, the three shown here are more "normal" than most I see in that they've got at least a semblance of a standard normal distribution bell curve in them -- kinda fattish in the middle, and tapering off toward each end. There are little spikey-guys at each end as well. Pretty typical, all in all.

Part of RAW conversion is doing a lot of it, learning to recognize what you've got to work with in a given photo, and making the best of that.

Anyway the three histograms shown have been adjusted by .1 Ev steps from -1.3 to -1.5 Ev -- and if you look not even too closely you can see that -1.4 Ev has the plumpest center section of the three.

(Further reductions or increases in Ev flattened-out the histogram pretty radically -- take my word for it -- see the "as-shot" histogram down below in the postscript).

So here's the image right before I clicked "OK" and "Convert" (clicky):


1.4_histogram

And here's the finished photo, after cropping and pulling up the low (black) end of the Levels by +5 (clicky):


3987_Vultures_v_Skyline_041310
Not necessarily the greatest Sports Photo of all Time, but a good example of Our Friend the Histogram(tm) in action.

p.s.:

1) Purists will wonder why I'm having to pull down the Exposure Compensation by -1.4 Ev in the first place.

Good question.

I shot this particular photo a little after sundown when I had just reset the EOS 5D Mark II's ISO from 1600 to 6400 in anticipation of the ongoing transition from fading daylight to full lighting by the field lights.

The original photo was pretty over-exposed. Here's the as-shot histogram:

as-shot_histogram

The histogram's as flat as a pancake, and all the data's run off the right (white) end. So by pulling down the Exposure Compensation I pull the data back toward the center of the histogram, and Compensate for the bad Exposure.

Cool, huh?

A little later (probably at half time) I would also switch the White Balance from Daylight to Fluorescent, which seems to be the best compromise for the color of the field lights at Vashon High School stadium.

2) The sharp-eyed will note that I'm saving as an 8-bit tiff.

The immediate project here was to produce tiffs for conversion to jpegs to put up on my web site. 8-bit depth is fine for something that's going to end up as a 600-pixel wide jpeg for web display.

Actually, if I was to get an order for a reprint off of this image, I'd go right back to the original RAW file and start from zero...