Color Temperature.....

[Daniel Browning]

Canon's page about sRAW doesn't really describe what it is. In reality, sRAW is actually a high quality 15-bit JPEG. In other words, the only difference between JPEG and sRAW is that the sRAW is 15-bit linear and JPEG is 8-bit log. The compression is the same as JPEG at the highest quality setting (where it only throws away half of the color information).


This is very different from true RAW, because it's already demosaic'ed -- all your raw converter can do is display the file (like a JPEG) -- it can't apply it's own demosaic engine. Personally, one thing I dislike about Canon's demosaic is that it causes mazing artifacts when noise is high, so I use other raw converters. But if I were to use sRAW, the artifacts would appear even with other raw converters.


That's not to say you shouldn't use sRAW. Most of the time, going from 15-bit JPEG ("sRAW") to 8-bit JPEG is worse than losing the resolution.

[jks_photo]

@ Daniel..... so basically what you

[Carlos Lindado]

I think it

[Daniel Browning]

I think it's important to mention that there's no such thing as a 15-bit JPEG,

Well, it may not be standards-compliant, but it does use the exact same "Lossless JPEG" file format as the embedded thumbnail, so I consider it JPEG. If you look at the source code to the raw converter, you'll see that it uses the exact same function to read the embedded thumbnail as it does to read the sRAW, the only difference is bit depth. They both even use the same 4:2:2 compression (i.e. throw away half the color information).


Perhaps a better way to think of it is as a 15-bit TIFF file with half the color pixels missing.

[Carlos Lindado]

If you look at the source code to the raw converter..

Are you referring to dcraw? (the raw reader used by Raw Therapee)

[Daniel Browning]

Yes.

[Carlos Lindado]

Well, that program in particular (dcraw) implements a lossless JPEG reading function (you can see Adobe

[Daniel Browning]

Right, it doesn

[Carlos Lindado]

Well, I think you're talking about the regular (lossy) JPEG standard, as the lossless JPEG standard doesn't use DCT, but DPCM (differential pulse code modulation). Besides, DCT and chroma subsampling aren't exclusive, indeed, they're used together to compress JPEG images (so it would be incorrect to state that chroma subsampling is the other type of JPEG compression besides DCT, since the two are used simultaneously).


OTOH, an sRAW file shouldn't be of the same size as a full RAW, since you're effectively discarding a lot of pixels and therefore a lot of bytes.

[Daniel Browning]

(so it would be incorrect to state that chroma subsampling is the other type of JPEG compression besides DCT, since the two are used simultaneously).

Well, the point is that sRAW uses chroma subsampling compression.



<meta http-equiv="content-type" content="text/html; charset=utf-8" />

OTOH, an sRAW file shouldn't be of the same size as a full RAW, since you're effectively discarding a lot of pixels and therefore a lot of bytes.

Not really. Canon starts with only a single color sample per pixel location, like every Bayer sensor, say, 21 MP. Uncompressed and without an embedded preview jpeg, that's 35 MB.Then Canon does a demosaic. This results in three color samples at every location -- so they just tripled the quantity of information -- now it's 105 MB (!) instead of 35 MB. (Almost as bit as 16-bit TIFF files from a full size raw conversion.)


So next they downsize to sRaw1, 10 MP. In doing so, they should (and do) increase bit depth to 15 bits to prevent quantization error. Uh oh, file size is 53 MB -- still more than the full-size raw. They *have* to apply chroma subsampling compression (throwing away color detail) just to get it down smaller than the original raw. 4:2:2 brings it down to 25 MB -- which is smaller than the 35 MB we started with. Then apply the standard huffman-coding lossless compression and add in a 3 MB preview JPEG and you're done.


It's truly sad that *this* is the best we can get from Canon. Nikon's so-called "lossy" NEF is far, far superior for the same file size.