The consolidated CRAW compression thread

I'm a bit surprised.

I thought that the only difference between Sony's uncompressed raw and compressed raw was the filesize and processing speed. In other words, I thought that cRaw was using a lossless compression.
After all, a100 is also using a lossless compression for its raw files.

Anyhow, I'd be interested by a test of this.

cRAW is most certainly lossy.
it uses a fixed length 8-bit compression of the 12-bit data.
"most" data in "most" cases should be preserved, but the compression scheme is complicated and can be lossy in certain cases.

If you are looking for proof that the scheme uses a fixed 8-bit compression or the scheme itself look at Dave Coffins dcraw.c
it has the code for this and will likely be used in a variety of converters.

vishnu

That's most interesting, so do we still get 12 bit data when it is uncompressed from cRAW?

I always wondered why Sony offers these 2 formats, and If I would have a A700 I would NOT shoot in cRAW.
It has to be something that is why it is called compressed and compressed for me doesn't sound good.

So this finally has given an answer.
Sounds very logical to me, so the Dynamic range in certain kind of images could be reduced even more in cRAW.

gm4jjj wrote: Be careful - bit depth and dynamic range are not the same thing.

Exactly !: Dynamic range is the difference between the darkest part that gives the smallest signal and the brightest part before it "blows out". Bit depth is how many levels this range between darkest and lightest is devided into.

Also be carefull to say that compression allways throws away data. There are compression algorithms which are scientifically proven to be lossless (eg. LZW compression of Tiff files).

Edited by pegelli - 09 November 2007 at 14:01

gm4jjj wrote: Sony's FAQ is very well crafted to avoid the answer as to whether it is lossy or not.

I can't be angry about this. Probably if they would word it differently a lot of people will discard it for the theoretical reasons mentioned in this and other threads while as you say in practice you cannot tell the difference. Probably even pixel peeping at high magnification won't show you what it does, so no way a print will be influenced

vnatchu wrote: If you are looking for proof that the scheme uses a fixed 8-bit compression or the scheme itself look at Dave Coffins dcraw.c it has the code for this and will likely be used in a variety of converters.

After a quick look at dcraw, cRaw is indeed lossy, and using 8 bits.
It seems that at each group of 16 bayer pixels, it stores the max value, and then the following 16 pixels will be stored using 8 bits with values that can go up to this max value.
So overall dynamic range is preserved, but you only have 8bits of dynamic per 16 pixels.

This is a very simple scheme, and for sure it could have been a bit more efficient, but overall it's not that bad.