Are compressibility checks really that important? [Archive]

CCoug83

14th January 2004, 07:13

I've been ripping for a while and it seems like compressability tests take up the most time in ripping. Honestly I've never really fully understood how they work or what exactly all the numbers mean.

I was just wondering if it really improves the quality of your video enough to make me want to do it all the time. I always make sure that the bits/(pixel*frame) is always above 0.2 however. That's why I'm a huge fan of AutoGK, but I miss the credit compression. So what's more worth it, using AutoGK and having the auto compressability check or sticking with GK with no compressability check and gaining the extra bitrate by having the credits compressed?

A good feature to add to the next GK version would be an auto compressability check feature such as AutoGK has. To me this is the most complex part of the ripping process, one I can't seem to master and would be much more comftorable having the computer do for me.

hakko504

14th January 2004, 08:57

The short answer is yes.

The long answer is still yes, but I'll also explain why:

A compression test finds the point where the codec won't assign any more bits to the video, i.e. where 100% is. That also means maximum quality, i.e. you can't get it to look better with these settings.
This is done by encoding short segments of the video from all over the video.
So what does a % say really?
If it's above 100% then you'll end up with a smaller video than you expected. So either you're encoding the video too small, applies too much filtering or you end up not using the whole CD that you wanted to archive the video on.
If you are very close to 100% then I suggest a single pass at Q2 instead of a multipass encode - you'll save a lot of time, but the video might end up with a slightly different size than you expected.
If the % is between 80 and 100 then you are getting as perfect as possible with the current settings - except that some versions of DivX failed to assign bits properly in this region and ended up with a slightly smaller video than expected.
If the % is between 60 and 80 then you're still in the green. You have found good settings, probably not filtering too hard (might be a problem for people with slow computers)
If the % is between 40 and 60% then you could probably do with some improvements like use a slightly lower resolution or some more filtering. If you want to use this then you should probably do a testclip first and see if you like the results before the main encode.
And finally, if the % is below 40 then you'll most likely be sorry if you encode the clip because you're not going to like the result. Depending on the source you'll end up with macroblocks, unsharp edges, etc. The codecs just can't cope with a video as complicated as this, and you'll have to sacrifice other things instead, see previous paragraph.

But now comes the paradox: If you lower the resolution then you remove details from the video as well as increasing the bpp value in order to make it easier to encode. But if you don't lower the resolution then the quality of the encode will be lowered resulting in worse looking video. So the question is: what looks best when re-scaled back to full screen for viewing? And the answer is: the low res video will win every time. Because when scaled back it will only introduce faults at pixels very close to the original source of the error (i.e. a line will be smeared a bit or just appear as unsharp) but the badly encoded hi-res video will move errors all over the screen, something the eye can't adjust for and thus finds it apalling.