The End of "95%"

95% compression is good only for material you shoot with a consumer or even prosumer camera (4;2;0 or 4;1;1) - that is, only some frames have both color and luminance information, while others have only luminance or partial color information. In that case, is it true the extra 5% adds "blank" information that does not exist.

100 % compression is good for 4;4;4 material - that is each and every frame has the full range of information: color and luminance. That happens to high range, proffesional cameras and to your rendered animations.

Unlike recorded camera clips, your rendered animations are 4;4;4 - that is each and every rendered frame of your animation has both color and luminance information. For your rendered animations use 100% in order to keep all your color information. 95% makes your clip much more smaller but your clip is loosing important color information. The compression level is easily viewable in the data rate. Shot material is about 36 megaBIT/sec for HD (SD is usually 25 mbits/s only). Rendered animations should go much higher. A professional buyer would look for animations with high data rate(much more color and details information).

This should be the end of the "95%" myth as a catholicon / panacea / cure-all / heal-all / all-heal.



Sorry for my poor English. I am hardly trying to learn Russian from old communist handbooks(as other are not available). This makes me loose my sense of good English.

God bless us all!
Honestly,
Christian I. Ionescu

Are you saying to do 100% then? This is only relevant to animations? what about about HD video...

As I said, only high end cameras use 4;4;4 and subsequently are worth 100% compression ratio.

If your camera does not shoot 4;4;4, then keep 95% compression ratio. As far as I know, Canon HV30 is 4;1;1 - so, stay with 95%.

Use 100% only for animations or camera like Red or a > 20k Sony.

Sorry :P

...but I disagree, compression is a method for throwing out data that isn't useful. For instance, in an area where a picture is black the jpeg throws out blue red green light data. Why? Because it doesn't make the black area look any better.

In compressed music (mp3s) the human ear can only hear the loudest sound at any given point in time. (The brain fills in the frequency gaps) ... So the data of other frequencies is thrown out because no one (human that is) can notice the difference.

Ive made CGI for B-movies that was made at 70% compression, and no one noticed.

All of the following images were compressed several times, can anyone really tell?
http://i43.tinypic.com/29m8175.jpg
http://i40.tinypic.com/2mfb2jb.jpg
http://i43.tinypic.com/5bc3t.jpg

My guess is Pretty eyes. Snow is a little grain look but well done due to dark sky and bright lights. Amalgamate is a true CG so looks sharp and therefore should more easily be compressed several times with little of no loss of quality. I can see all are good but can also see why each could easily have been compressed numerous times.

It does depends on the algorithm used in compression, videos that are heavily compressed for streaming video are noticable and blocky. However these are jpeg.

I believe a video should be compressed as much as possible but not interfer with the look or the beauty of the clip/picture. I personally don't think the compression interfers with the beauty of these pictures.

A side note....I know this moderately wealthy guy who wants to be a director( he has no experience BTW), he bought all this really expensive video equipment, multiple p2 cameras, cranes (the big kind), all kinds of lights. He is always concerned about compression artifacts, he does a million white balances every 5 minutes, and he always wants to preserve the "quality" of his images.... but everything he shoots is boring, no concern for lighting, or acting quality. His productions suck! lol! I try to help him, but he only wants 'yes men' to go along with his ideas, he can't get a distribution deal, even with the worst distribution companies on the planet.

I would be shocked as hell if anyone could tell a 100% JPEG Animation and a 90% apart just by looking at them.

The difference between 70 or 95% and 100% is the extra channel of color:
- if you have a 4;4;4 signal, the 100% compression will keep all the channels
- if you have a 4;2;0 or 4;1;1 signal, the 100% compression will transform it in 4;4;4 signal adding blank(non existing) information without improving quality, only rising the file size.

@bryanbush: The difference between 100 and 95% is not only visually. The visual difference between 100 and 95% depends on the original file. If one starts from an uncompressed file(let's say a rendered animation) there might be visible differences regarding small details and gradients. Also, the compression quality strongly affects the keying quality. Try and make a chroma on a 4;2;0 or a 4;1;1 or a 95% compressed signal.

@Iron: this is not a speak on stills. The jpeg examples are irrelevant. Which frame would you take as example from a 4;2;0 or from a 4;1;1 signal? Only in a 4;4;4 signal all the frames are the same(speaking of information)

I'm not talking about for keying, but really with a black and white alpha mat you can do more with less in this instance you don't need a 16 bit black and white only an 8 bit, if your rendering with + Alpha or Alpha in the image it's going to be nice and large. Testing it when you do compression is a must with + Alpha depending on it's final output. Like IronStrike said for the B movie it did not matter 70% worked perfectly.

Also what I'm saying is with the naked eye your not going to see a visual quality loss at 90% I think the end user would rather the file size be a faster download then have some thing that they cant even see. If your getting banding at 90% or having issues that you can see by all means crank it up... Artifacts, banding, jagged edges, none of it looks good.

Stills is the same as animation we deal in frames so his still examples are perfect for what we are talking about and that is visual quality degradation.

For footage and animations aimed at heavy compositing and keying Christian (ionescu) is absoultly correct. 100% is the way to go (if the original signal has 4:4:4 colour coding).

However, if you're doing generic backgrounds or generally stand alond type of shots, 100% would be entirely overkill.

Compression is always a type of trade-off. Quality against size. Use it wisely. There is no such thing as a heal-all, correct for everyone solution.