Björn Benckert
Well-known member
Your use of the term "throw out" is indicative of the way a lot of people look at compression, and it is very inaccurate. Information is never "thrown out." What compression techniques do, among other things, is look for things like redundancies in the data to allow a lot of information to be expressed in a much smaller space. So, for instance, if you photograph a black table, and the light is fairly even, you can express those values digitally in multiple ways. One way would be to write every pixel into a file with the same value, so you have a one to one relationship - i.e., if you're describing 1000 pixels, you have 1000 recorded values. Or, you could say "take the first value and use it for the next 1000 pixels." The second way takes up about 999 times less space in a file, but when the image is reconstructed, you get the same result. That's what would be referred to as "lossless" compression. Other techniques use mathematical formulas to record information in a similar way (using channel math, things like subtracting the green value from the red and the blue to produce smaller numbers to encode, are common). Still others use different levels of resolution to express the image at different detail levels which, when combined, yield a very close match to the original image, but can also provide less detailed versions at lower resolutions by not including the "overlays" of the highest resolution, and only extracting the lower resolution records. That's essentially what wavelet compression is, and that is the basis for what Red - and Sony, and Cineform, and others - do. So while you might be under the impression that information is simply "thrown away," that is not the case. It is encoded in a way that takes up less space, and the amount of precision that is retained is dependent on both the level and type of compression used. Obviously, I'm leaving out a lot of other details (intra vs. interframe encoding, DCT algorithms, etc., etc.) but that is the gist of it. The Cinema DNG format was developed by Adobe as an adjunct to its DNG format that was already on the table for still images, and proposed as a standard, but it has only really been adopted by Blackmagic, Ikonoscop, and Adobe at this point. As I recall (and I could be wrong on this) the original format allowed for a degree of lossless compression, and that is likely what Blackmagic is employing. I'm not aware of the specific techniques they're using, but suffice it to say that what Red does with Redcode and what Blackmagic does are considerably different, with different results. Redcode is lossy, but with the higher resolution images that Red captures that is far less significant. Blackmagic's is a lot less lossy - and possibly lossless - but that also limits the amount of compression that they can use, which explains why their files are considerably larger even though the captured image is at a lower resolution. In terms of color grading, none of this is particularly significant because you can't grade any image until you decompress it. Anything that you're doing is being done on a fully reconstructed RGB image by definition. So although one needs to be aware of possible artifacts generated by either the compression, decompression, or reconstruction processes, it has no direct bearing on the colors you mix.
Hopefully that doesn't complicate this even more.....
jupp that is pretty much how it works.
graded a BM image today, must say I kind of like the feel of the image. I can push things around quite far without seeing any sort of "digital artifacts"
Here is source (as jpg): http://www.syndicate.se/Files/~usr/hawk/oddgroup_1_2013-09-03_2246_C0014_000222.jpg
and here is the same frame that past tough flame: http://www.syndicate.se/Files/~usr/hawk/KodakPrint_0002greeboost.jpg
I added the noise, not becuse I needed to hide artifacts, more since I though it suited the shot.
Question anyone that has calculated how much data loss epic 3:1 is. as I understand the "3:1" number is more a ratio for file size compared to uncompressed / unpacked footage. But how much of the 2 missing thirds are actually taken out of the equation and how many are packed in... do not think I use the right terminology here but I think you understand what I mean.