Raw File Compression.
Today I’m going to give you my point of view over that most vexatious question – is LOSSLESS raw file compression TRULY lossless?
I’m going to upset one heck of a lot of people here, and my chances of Canon letting me have any new kit to test are going to disappear over the horizon at a great rate of knots, but I feel compelled to post!
What prompts me to commit this act of potential suicide?
It’s this shot from my recent trip to Norway:
Direct from Camera
Processed in Lightroom
I had originally intended to shoot Nikon on this trip using a hire 400mm f2.8, but right at the last minute there was a problem with the lens that couldn’t be sorted out in time, so Calumet supplied me with a 1DX and a 200-400 f4 to basically get me out of a sticky situation.
As you should all know by now, the only problems I have with Canon cameras are their short Dynamic Range, and Canons steadfast refusal to allow for uncompressed raw recording.
The less experienced shooter/processor might look at the shot “ex camera” and be disappointed – it looks like crap, with far too much contrast, overly dark shadows and near-blown highlights.
Shot on Nikon the same image would look more in keeping with the processed version IF SHOT using the uncompressed raw option, which is something I always do without fail; and the extra 3/4 stop dynamic range of the D4 would make a world of difference too.
Would the AF have done as good a job – who knows!
The lighting in the shot is epic from a visual PoV, but bad from a camera exposure one. A wider dynamic range and zero raw compression on my Nikon D4 would allow me to have a little more ‘cavalier attitude’ to lighting scenarios like this – usually I’d shoot with +2/3Ev permanently dialled into the camera. Overall the extra dynamic range would give me less contrast, and I’d have more highlight detail and less need to bump up the shadow areas in post.
In other words processing would be easier, faster and a lot less convoluted.
But I can’t stress enough just how much detrimental difference LOSSLESS raw file compression CAN SOMETIMES make to a shot.
Now there is a lot – and I mean A LOT – of opinionated garbage written all over the internet on various forums etc about lossless raw file compression, and it drives me nuts. Some say it’s bad, most say it makes no difference – and both camps are WRONG!
Sometimes there is NO visual difference between UNCOMPRESSED and LOSSLESS, and sometimes there IS. It all depends on the lighting and the nature of the scene/subject colours and how they interact with said lighting.
The main problem with the ‘it makes no difference’ camp is that they never substantiate their claims; and if they are Canon shooters they can’t – because they can’t produce an image with zero raw file compression to compare their standard lossless CR2 files to!
So I’ve come up with a way of illustrating visually the differences between various levels of raw file compression on Nikon using the D800E and Photoshop.
But before we ‘get to it’ let’s firstly refresh your understanding. A camera raw file is basically a gamma 1.0, or LINEAR gamma file:
Linear (top) vs Encoded Gamma
The right hand 50% of the linear gamma gradient represents the brightest whole stop of exposure – that’s one heck of a lot of potential for recording subtle highlight detail in a raw file.
It also represents the area of tonal range that is frequently most effected by any form of raw file compression.
Neither Nikon or Canon will reveal to the world the algorithm-based methods they use for lossless or lossy raw file compression, but it usually works by a process of ‘Bayer Binning’.
If we take a 2×2 block, it contains 2 green, 1 red and 1 blue photosite photon value – if we average the green value and then interpolate new values for red and blue output we will successfully compress the raw file. But the data will be ‘faux’ data, not real data.
The other method we could use is to compress the tonal values in that brightest stop of recorded highlight tone – which is massive don’t forget – but this will result in a ’rounding up or down’ of certain bright tonal values thus potentially reducing some of the more subtle highlight details.
We could also use some variant of the same type of algorithm to ‘rationalise’ shadow detail as well – with pretty much the same result.
In the face of Nikon and Canons refusal to divulge their methodologies behind raw file compression, especially lossless, we can only guess what is actually happening.
I read somewhere that with lossless raw file compression the compression algorithms leave a trace instruction about what they have done and where they’ve done it in order that a raw handler programme such as Lightroom can actually ‘undo’ the compression effects – that sounds like a recipe for disaster if you ask me!
Personally I neither know nor do I care – I know that lossless raw file compression CAN be detrimental to images shot under certain conditions, and here’s the proof – of a fashion:
Let’s look at the following files:
Image 1: 14 bit UNCOMPRESSED
Image 2: 14 bit UNCOMPRESSED
Image 3: 14 bit LOSSLESS compression
Image 4: 14 bit LOSSY compression
Image 5: 12 bit UNCOMPRESSED
Yes, there are 2 files which are identical, that is 14 bit uncompressed – and there’s a reason for that which will become apparent in a minute.
First, some basic Photoshop ‘stuff’. If I open TWO images in Photoshop as separate layers in the same document, and change the blend mode of the top layer to DIFFERENCE I can then see the differences between the two ‘images’. It’s not a perfect way of proving my point because of the phenomenon of photon flux.
Photon Flux Andy??? WTF is that?
Well, here’s where shooting two identical 14 bit uncompressed files comes in – they themselves are NOT identical!:
The result of overlaying the two identical uncompressed raw files (above left) – it looks almost black all over indicating that the two shots are indeed pretty much the same in every pixel. But if I amplify the image with a levels layer (above right) you can see the differences more clearly.
So there you have it – Photon Flux! The difference between two 14 bit UNCOMPRESSED raw files shot at the same time, same ISO, shutter speed AND with a FULLY MANUAL APERTURE. The only difference between the two shots is the ratio and number of photons striking the subject and being reflected into the lens.
The Levels Adjustment Layer had values of 0, 50 and 150, and is going to be added to the following comparison images:
Firstly 14 Bit UNCOMPRESSED compared to 14 bit LOSSLESS (the important one!):
14 bit UNCOMPRESSED vs 14 bit LOSSLESS
Please remember, the above ‘difference’ image contains photon flux variations too, but if you look carefully you will see greater differences than in the ‘flux only’ image above.
The two images above illustrate the differences between 14 bit uncompressed and 14 bit LOSSY compression (left) and 14 bit UNCOMPRESSED and 12 bit UNCOMPRESSED (right) just for good measure!
As I indicated earlier in the post, this is not a definitive testing method, sequential shots will always contain a photon flux variation that ‘pollutes’ the ‘difference’ image.
I purposefully chose this white subject with textured aluminium fittings and a blackish LED screen because the majority of sensor response will lie in that brightest gamma 1.0 stop.
The exposure was a constant +1EV, 1/30th @ f 18 and 100 ISO – nearly maximum dynamic range for the D800E, and f18 was set manually to avoid any aperture flicker caused by auto stop down.
You can see from all the ‘difference’ images that the part of the subject that seems to suffer the most is the aluminium part, not the white areas. The aluminium has a stippled texture causing a myriad of small specular highlights – brighter than the white parts of the subject.
What would 14 bit uncompressed minus 14 bit lossless minus photon flux look like? In a perfect world I’d be able to show you accurately, but we don’t live in one of those so I can’t!
We can try it using the flux shot from earlier:
But this is wildly inaccurate as the flux component is not pertinent to the photons at the actual time the lossless compression shot was taken. But the fact that you CAN see an image does HINT that there is a real difference between UNCOMPRESSED and LOSSLESS compression – in certain circumstances at least.
If you have never used a camera that offers the zero raw file compression option then basically what you’ve never had you never miss. But as a Nikon shooter I shoot uncompressed all the time – 90% of the time I don’t need to, but it just saves me having to remember something when I do need the option.
Would this 1DX shot be served any better through UNCOMPRESSED raw recording? Most likely NO – why? Low Dynamic Range caused in the main by flat low contrast lighting means no deep dark shadows and nothing approaching a highlight.
I don’t see it as a costly option in terms of buffer capacity or on-board storage, and when it comes to processing I would much rather have a surfeit of sensor data rather than a lack of it – no matter how small that deficit might be.
Lossless raw file compression has NO positive effect on your images, and it’s sole purpose in life is to allow you to fit more shots on the storage media – that’s it pure and simple. If you have the option to shoot uncompressed then do so, and buy a bigger card!
What pisses my off about Canon is that it would only take, I’m sure, a firmware upgrade to give the 1DX et al the ability to record with zero raw file compression – and, whether needed or not, it would stop miserable grumpy gits like me banging on about it!