Khaki F. Dog

After posting yesterday’s article extolling the virtues of the new Lossy Digital Negative format that Adobe’s introducing in the upcoming Lightroom 4, I re-read this article, which pointed out a conceptual gaffe that really ought to have occurred to me.

Background: camera sensors count their resolution in megapixels, but that isn’t totally fair. Modern camera sensors (except for Sigma’s expensive Foveon sensors) use a Bayer matrix, where each light-sensitive element (pixel) is only sensitive to either Red, Green or Blue.

When your camera saves a RAW file, it therefore saves a sequence of Red, Green and Blue values. To view this as a full-color image it needs to be de-mosaiced, a process whereby each pixel’s total color value is created by 33% fact and 67% guesswork. For each Red pixel, the demosaicing algorithm looks at the neighboring Green and Blue pixels and guesses roughly how much Green and Blue would have landed on the Red pixel.

The Foveon sensors I mentioned use an innovative technology where each pixel captures all three colors, and Sigma has a terribly hard time explaining why their megapixels are actually three times as accurate as everybody else’s megapixels.

So, okay, that’s the reality. An actual RAW file only records one third of the data, with each pixel containing either R, G or B data. Fine.

But when you save as Lossy DNG, each pixel is recorded as full RGB data. Two-thirds of the data you’re saving in a Lossy DNG is guesswork, it’s data that wasn’t actually captured by the sensor.

There are performance benefits, obviously. By storing each pixel’s full RGB data the file can be loaded quicker, because the computationally complex guesswork has already been done and you don’t have to wait for the image to be demosaiced.

The whole point of Lossy DNG, though, is space saving.

Un-demosaiced image data presents its own problems, of course, and no doubt there are significant engineering hurdles to overcome when trying to compress fragmented pixel data, so it’s likely Adobe made the right call.

Still, bit of a bummer.

Photography nerdery ahead. Be warned!

Some years ago Adobe introduced the Digital Negative Format, or DNG, as a universal RAW format for photography.

A quick primer. JPEG files store 8 bits of image data in each channel, so a JPEG file can describe 16.7 million colors. This is beyond the capacity of the human eye to distinguish, or the capacity of most screens to reproduce. However, it’s only a fraction of what the human eye can observe. Sunlight can be one hundred thousand times brighter than candle-light, but we can still see clearly, and distinguish millions of colors, under both circumstances.

RAW files capture all the sensor’s data, which can capture much more than the 8 bits stored in the resulting JPEG. And by ‘much more’, I mean billions of times more data. See here, in this example. Because the RAW file from my Nikon stores over 280 000 000 000 000 colors, I could brighten the shadows and darken the highlights and pull out perfectly pretty colors. Score.

Each manufacturer uses their own special RAW format to suit their hardware, so DNG was an effort to standardize RAW formats. A DNG file contains the RAW data, plus the manufacturer’s interpretation algorithms, and reduces the file size without any quality loss. Future-proof, flexible and compact.

In the upcoming Lightroom 4, Adobe introduces a new spec for DNG. One of the new features is Fast Load Data, which should speed up loading the (often very large) DNG files, though my experience shows this benefit to be negligible. The more dramatic new feature is the option of ‘Lossy’ DNG.

It took me some research to figure out exactly how this new option would work (this thread was particularly helpful) but it seems to actually be really, really clever.

Lossy DNG is not RAW. It uses JPEG compression algorithms, but applies these in such a way that the resulting file is very close to a full-blooded RAW file in terms of flexibility. For each individual channel, it stores 8 bits worth of data, but it also records which 8 bits.

If you were to shoot a very dark photo and record the scene as RAW and JPEG, which are capable of recording up to 281 trillion and 16.7 million colors respectively, you’d only use a fraction of the colors they’re capable of recording. Say, the darkest 5% of the possible colors. That means the RAW file will still capture 14 trillion colors, but the JPEG will capture only 800 000 colors, significantly less than the limit of the human eye’s ability to distinguish.

In the example earlier, for instance, the grass is completely dark, so out of all the colors my RAW file can store, probably only the darkest ones are actually included in this image. When I save this as a Lossy DNG, Lightroom is smart enough to record only the colors that actually occur, convert those to 8 bit, and record some code that effectively says “the data recorded here is from the darkest 5% of the original RAW file”.

Still, practically, a Lossy DNG is really close to a full-blooded RAW file in terms of editability. Because each channel is processed separately you can easily apply radical shifts in white balance, which is all but impossible with JPEGs. I actually saved a copy of the example’s RAW file as a Lossy DNG and reapplied all the edits I’d made to the original. I was hard-pressed to find any differences in the results on my screen.

And still, a Lossy DNG is often 75% smaller in terms of file size.

Storage space today is cheap, but not free. While I wouldn’t dream of sacrificing the tiniest bit of fidelity from the wedding and baby photos I’ve shot, I’m much less worried about photos I took of outings with friends. In fact, I’d be fine with just converting those to JPEG, if that were easy to do.

Much smaller files, while still maintaining almost all the flexibility of full-blooded RAW? Golden middle road, for casual photos at least. The biggest drawback right now is that only the upcoming Lightroom 4 can read Lossy DNGs, but other applications will surely catch up soon enough.

Good job, Adobe.