why would I convert to 16 bits TIFF ?

[dmalek]

Hello everyone,

Maybe be this topic has alreayd been discussed?

here is my question : i own an EOS10D. A CRW file is about 6 Mb in size. Starting from the 6 Mb file i can get a 20 Mb 8bits TIFF file or a 16 bits TIFF file (38 Mb). Why would I use the 16 bits TIFF? How could 6 Mb of info provide enough details for a 40 Mb file ?

thank you in advance ;)

[vvizard]

I don't know enough about 8 vs 16-bit, to answer that, but the reason a 6MB *.crw file holds enough info, is that it's compressed lossless. Opposite of how jpeg works, a lossless compression doesn't discard anything when packing. If you pack something lossless, you will get the exact same result when unpacking. If you're familiar with *.zip or *.gz, that's how lossless work. You wouldn't want to pack your programs in a format (jpeg) that discards info, and cut corners ;) You depend on that whatever you pack, it will be identical when you unpack it, and that's how the EOS-models do their compression. Hope this helps

[robertwgross]

Yes, the RAW CRW file is about 7MB, roughly (6 to 9). That is 12-bits per pixel color channel encoded, or it could be 36 bits total. If you want to, you can use the best 8 bits per color channel (24-bit TIF), or you can use 16 bits per color channel (48-bit TIF).

I convert to 48-bit TIF, archive those, and then down-convert to 24-bit TIF so that it looks the same as the original 48-bit TIF. Then any refinement of the image happens there in 24-bit land.

Disk space is cheap!

---Bob Gross---

[NickSI]

This actually two different questions :)

Why would I use the 16 bits TIFF?

In raw format you got 12-bits per color channel. TIFF format can keep either 8 or 16. More bits means that you can define pixel color with greater accuracy so you get smoother halftones.

Unfortunately, only few image editors can work with 16-bit image and amount of operations available is limited (Adobe promise to fix this situation in PS 8 ).

I usually keep originals in RAW formats, convert them to 16-bits, do some processing and save final results in 8-bit format.

How could 6 Mb of info provide enough details for a 40 Mb file ?
OK. RAW file is actually black and white with only luminosity reading of each CMOS cell stored. So for each pixel you have to keep 12-bits of data, not 8x3=24 (8bit TIFF)or 16x3=48 (16-bit TIFF). It is up to build-in camera CPU or you RAW conversion software to link this data to the Bayer filter pattern, which sits on the top of CMOS matrix, and calculate proper colors.
Also, as CRW is proprietary format, they can keep exactly 12-bit per pixel without padding them to 16-bit in TIFF (another 25% space savings).
An finally, Canon probably applies some loose less compression in order to shrink it even more.

[Webster]

OK. RAW file is actually black and white with only luminosity reading of each CMOS cell stored. So for each pixel you have to keep 12-bits of data, not 8x3=24 (8bit TIFF)or 16x3=48 (16-bit TIFF). It is up to build-in camera CPU or you RAW conversion software to link this data to the Bayer filter pattern, which sits on the top of CMOS matrix, and calculate proper colors.


This statement is almost right, but also misleading. It is correct to say that each CMOS cell only stores luminousity, but to go on and say that this makes the RAW file B&W is errroneous. The RAW file isn't any color at all, it's just a bunch of numbers that, when correctly combined with the Bayer filter pattern can be turned into some other, more standard bunch of numbers than can then be processed by hardware and software and made into an image.

Also, while it is true that each cell just holds luminousity, it takes more than one cell to make up a single pixel. It's not the very neat three cells that it would take if it were RGB - the Bayer pattern is much more complex than that. But I don't think much, if any, of the file size saving is gained by the fact that cells only record luminousity.

[jimmyd]

my understanding is this: if what your'e shooting is going to a graphics house, you would not convert to 16 bits as they cannot use it. 8 bits is what they want. I don't know if that solves your dilemna

[robertwgross]

Typical run-of-the-mill graphics houses will find 24-bit TIF files or JPEG files to be plenty fine to work with. You can, however, get just a little more Zing (that is a technical term :-) out of a shot in 48-bits total. This is expecially true if you need to make any major dynamic scale changes and fine color tuning. If you are just going to shoot, process, and run to the nearest inkjet printer, then 24 bits is normally sufficient.

---Bob Gross---

[Dave I]

webster wrote:
But I don't think much, if any, of the file size saving is gained by the fact that cells only record luminousity.

Sure it is. The RAW file only saves 12 bits per photosite. After RAW -> RGB TIFF conversion, each photosite's data is stored as a RGB pixel of 24 (8R + 8G + 8B) or 48 (16R + 16G + 16B) bits. That's a savings in storage requirements of 1/2 or 3/4 (only 1/2 or 1/4 of the number of bits need to be stored per photosite).

If there were such a thing as a 12-bit TIFF file, you'd still only have to store 12 bits per final pixel location in RAW vs 36 in the (hypothetical) 12-bit TIFF. That's 2/3 savings before compression

Also, while it is true that each cell just holds luminousity, it takes more than one cell to make up a single pixel. It's not the very neat three cells that it would take if it were RGB - the Bayer pattern is much more complex than that.

I'd like to clarify that there is one photosite ('cell') per pixel in the final image. Your statement kinda makes it sound like there are more photosites than final image pixels. This is assuming standard Bayer interpolated sensors, not exotic stuff like Foveon, of course.

[dmalek]

thank you all for your comments, they answer my questions and this is a very interesting information i'm going to forward to my friends ;)

[dbarthel]

When you adjust the image with curves or whatever, 16 bit tiffs prevent gaps in the histogram. Severe gaps can cause printer banding. So try to do as much image adjustment in 16 bit, including scaling, before converting to 8 bit, sharpening, and printing. In most cases with modest adjustments, you won't see much if any difference, but there are times it really makes a difference. I always shoot raw, and always archive the raw file, as it is a digital negative. You can convert the raw file in as many ways as often as you want, without losing your negative.

Dan