Raw compression won't be apparent in every image, but there are circumstances in which it risks undermining what should be a great image. Photo by Rishi Sanyal

Update: September 15, 2015: Sony has announced that the a7S II will have the option to shoot uncompressed Raw, with firmware being rolled out for some existing models, including the a7R II used here, later.

A Raw file is a Raw file, right? Well, not exactly. Lately, there's been a lot of talk (and a lot of anger) about the compression Sony uses in its Raw files. Compressed Raw files aren't uncommon, but they're usually compressed in a way that retains all the original 'raw' data from the sensor.

Instead Sony has, for several years now, chosen to apply non-optional lossy compression to its Raw output. This isn't likely to be an issue for many users, since the impact is generally quite small, but as the company shows its ambitions in the high end market, with the launch of the a7 series and a dedicated support system for professionals, the impact on image quality deserves a little scrutiny.

What exactly is going on with Sony's Raw files, and what might the potential impact be? The compression system has been investigated and detailed by Iliah Borg and Alex Tutubalin, and we've tried to distill their findings. It's important to keep this in perspective, though: in many circumstances you won't see this impact or encounter the limitations it can impose. 

What's happening?

Sony's compression process has two parts, and each of the two aspects have a different impact on the Raw files. The first step applies a compression curve to the data. This is a bit like a tone curve and is used to map the 14 bits of captured data down to an 11-bit space.

Although this part of the process is lossy, a well-designed compression curve has very little impact on image quality. This is because although shot noise makes up a decreasing proportion of the captured data in bright regions, the actual magnitude of the noise increases. This means it doesn't make sense to retain all the information about bright regions of the images, since a lot of that information will just be recording the subtleties of the noise.

Ideally, then, you can get away with a lot of compression in bright tones so long as you preserve increasing amounts of information at the dark end of the file. Unlike the (optional) compression curve Nikon uses, Sony's doesn't fully exploit this phenomenon, meaning that some useful data is lost, as well as some of the noise. This will, theoretically, reduce the dynamic range available in the files.

Stage two

Sony's Raw compression then has a second stage, where the image is divided up into a series of 16 pixel stripes for each color channel. Rather than recording a separate value for each of these pixels, the Sony system records the brightest and darkest value in each stripe, and a series of simple notes about how all the other pixels vary from those extremes. These notes are recorded using fewer bits than it would take to record the actual pixel values and it's this step that appears to cause most of the problems.

When there's not much difference between the brightest and darkest pixel, the system is able to describe the scene pretty well. However, as soon as you have a big gap between bright and dark, then the 7-bit values used to note the differences of the individual pixels aren't sufficient to precisely describe the original image information.

The localized, second step of the compression means detail around high-contrast edges isn't correctly recorded and this inaccuracy becomes increasingly apparent if you try to adjust those regions of an image.

This image was created from a Raw file that had been underexposed, to retain highlight detail, then pushed 5EV when it was processed. This may seem extreme but it can be useful to be able to selectively brighten areas of an image, without the risk of revealing lost data.

This imprecise recording of the original image data leads to artifacts in stripes around high-contrast edges in the photos. These errors can become even more pronounced if you increase the brightness or contrast in those regions when processing the files.

14-bit and 12-bit readout

To compound matters, several Sony cameras we've tested appear to switch their sensor read-out from 14 bits to 12 bits in certain modes: further reducing the amount of dynamic range that the camera ever captures, even before the effects of the compression process are brought into play. Continuous shooting, Bracketing and Bulb exposure modes will all push most Sonys down to 12-bit capture mode, which is then subjected to the two stage compression.

Conclusion

These approaches probably made sense in consumer-grade cameras, back in the days where processing power and storage space came at a significant premium. However, on a camera as expensive as the a7R II, which is likely to be used for quality critical shooting, it's hard to justify clumsy compression that can, depending on the image, throw away data you were expecting to have access to.

Overall, the effects of this compression aren't often visually significant. Their impact should mostly be understood as a reduction in processing latitude, since it tends only to be when you push and pull the Raw files that the missing data becomes visible. The compression curve throws away more shadow data than would be ideal: reducing dynamic range. There's a further reduction if you shoot in a mode that drops the camera into 12-bit readout mode. Meanwhile, the localized compression of tonal differences only has an impact near high-contrast edges.

And it's not just heavily-pushed images that start to reveal the lost data: this is a straight out-of-camera JPEG file (shot with DRO Auto), still showing stripes extending from a high-contrast edge. It looks still worse in a gently processed Raw conversion.

It's worth noting, though, that the impact of this process is lessened in the company's higher pixel-count bodies, since a 32-pixel stripe will be a smaller proportion of the a7R II's 40MP images than it would in a 12MP image from the a7S.

It's quite possible to shoot for years and never notice the impact of these design choices Sony has made, but they do add up to mean that you can't access the full capability of the camera. Most people shoot Raw precisely because they want to preserve the maximum possible processing latitude and keep their creative options open. Raw compression isn't the end of the world by any means, but it throws away a little bit of a camera's capability, which might be a little hard to swallow if you've paid multiple thousands of dollars for a cutting-edge camera.


This article is based on the investigative work of Alex Tutubalin and Iliah Borg, creators of LibRaw and Raw Digger.