Video Compression as Fast As Possible

Techquickie ·Techquickie ·2015-05-07 · 1,178 words · ~5 min read
Floatplane YouTube

Transcript

JSON SRT VTT 89
0:00 video compression is a very important aspect of how video streaming websites
0:03 like youtube netflix and hulu work without it you wouldn't be able to
0:07 stream such high quality high frame videos like this one first some math
0:11 let's say you're watching a video with a resolution of
0:15 pixels that's like 2 million pixels per frame and in this case you've got 30
0:20 frames per second that's a whopping 62
0:23 million pixels per second if each pixel needs 24 bits or three bytes of
0:28 information you're looking at 178 megabytes of data per second completely
0:34 uncompressed that would be 51 gigabytes of information for a single five minute
0:39 video we've got pretty fast internet here at the office but even we can only
0:43 download like two and a half gigabytes in five minutes
0:46 that's a long ways away from 51. fortunately a five minute high
0:50 definition youtube video is not 51 gigabytes it's in fact more like 72
0:55 megabytes that's like 700 times smaller
0:58 how did they do that that's the magic of video compression which works mostly by
1:03 minimizing redundancy in the video data
1:07 now there are all sorts of different ways to compress video and lots of
1:10 different video containers and codecs but i won't go into detail about that
1:14 because last already covered it in this episode and frankly i still don't really
1:18 understand it all even though i am employed as a video editor so don't tell
1:22 Linus so here's what you need to know there are two ways to compress video
1:27 spatial compression and temporal compression also known as intra-frame
1:32 and interframe respectively seriously who names this stuff
1:36 anyway i'm going to describe each of these in terms of the jpeg image format
1:40 and the mpeg video format spatial or
1:43 intra-frame compression is applied only to individual video frames to compress a
1:48 video frame you can use the same process that's used to compress a still image
1:52 like a jpeg when a jpeg is created color
1:55 information of the image is reduced in a process called chroma sub sampling and
1:59 then the image is split into sections of 8x8 pixels called macroblocks then all
2:04 of this fancy stuff is done to these blocks to further reduce the file size
2:08 it's easy to see this effect on an image that has been saved with a high amount
2:12 of compression all those ugly looking squares are known as compression
2:16 artifacts some images like this red square can be compressed more than
2:20 others like this complicated photograph because the red square is very simple
2:24 with a lot more redundancy so that's how a jpeg uses spatial compression now for
2:29 an mpeg sometimes called a motion jpeg
2:32 things are a bit more complicated than just lining up a whole bunch of jpegs
2:36 one after the other to be used as video frames doing it that way would not allow
2:40 for inter frame or temporal compression
2:44 temporal compression works again by reducing redundancy if you go through a
2:49 video like this one frame by frame you'll notice that many frames are
2:52 almost completely identical that's redundancy which can be easily
2:56 compressed for example for all the parts of this video that don't need to change
3:00 from one frame to the next like all this whiteness around me all we need is an
3:04 instruction for those parts of the frame which says don't change anything like
3:08 jpeg the mpeg standard breaks a video frame into 8x8 pixel macro blocks and
3:14 each macro block receives instructions on what to do with the pixels they
3:17 already have there are instructions for staying exactly the same for moving
3:22 rotating changing color changing completely and so on video frames with
3:27 instructions like this are called p frames and they use about half as much
3:31 data as an iframe which is pretty much just a jpeg
3:35 there are also b frames which are predictions or interpolations between i
3:40 and p frames b frames use a quarter as much data as an iframe so they really
3:44 save a lot of space this is all pretty easy to see on a video file that has
3:48 been saved in an extremely low quality setting like what you're seeing right
3:52 now again all these awful square thingies are called compression
3:56 artifacts furthermore sometimes a video file will be corrupted or missing some
4:00 data when that happens you get an effect like this one
4:04 this particular effect is caused by a missing iframe which would have cleared
4:08 away the old image to make room for a new one without the iframe the list of
4:12 changes in the following b and p frames are applied to the wrong image so now
4:17 you know why a video sometimes looks weird or gets all blocky those artifacts
4:21 are a result of the video compression being either too high or else some kind
4:26 of error in the coding or transmission of the video data now listen if none of
4:30 that made any sense everything i've said can be broken down to one thing
4:35 bit rate the bit rate is the amount of data or bits that are being used every
4:40 second if a video has a low bid rate it will be low resolution and or low
4:44 quality with lots of compression artifacts if a video has a high bitrate
4:49 it will be high quality and or high resolution so the more bits you have to
4:54 work with the better the video will look there's just no getting around that even
4:58 with the best compression methods video compression is a balancing act between a
5:02 good looking video or a small file size you can't have both but if compromise is
5:07 not your style maybe you'll like fractal design now i'm sure you've all grown
5:12 accustomed to embarrassing torturous events happening to line us in these
5:15 spots let's look at them now
5:20 behind in this manner
5:24 speaking of mean fractal design is that buying up all of my advertising
5:27 inventory so i was a little nervous about doing this ad i am however
5:32 relieved to say that josh has granted me a reprieve i guess Linus is the one he
5:36 wants to torture so we're using this spot to thank all of you the viewers who
5:40 went to the survey page and responded so positively to our previous fractal
5:44 design ads as fast as possible that they've decided to continue sponsoring
5:48 this show so thanks to you guys and thanks to fractal design the company
5:52 whose cases power supplies and fans bring all of the boys to all of the
5:56 yards be sure to check back over the next few weeks to learn just what that
6:00 reference might mean thanks for watching this episode of fastest possible give us
6:04 a like or a dislike leave a comment subscribe do it subscribe and we'll see
6:09 you next time