How Video Compression Works

[00:00] Have you ever thought about how video streaming

[00:03] Let's think about how big a typical 1080p

[00:10] 30 frames per second...

[00:11] That's almost 1.5 gigabits per second.

[00:14] How can you transmit that much data, over

[00:18] The answer is video compression.

[00:21] You might have heard of codecs.

[00:23] A codec is a piece of software that encodes

[00:27] The encoding part compresses the data, making

[00:32] And the decoding part reverses this process,

[00:37] possible.

[00:38] Use of codecs is not limited to video, they

[00:43] of signals.

[00:44] But for now, let's focus on how video codecs

[00:46] In the previous video, we talked about how

[00:50] In short, images are compressed by throwing

[00:54] the human eye and storing redundant data more

[00:58] We can easily extend image compression to

[01:03] by frame.

[01:04] This approach is called spatial or intra-frame

[01:07] Even doing that alone would significantly

[01:12] do much more than that.

[01:14] In a typical video, many consecutive frames

[01:18] We can make use of this temporal, inter-frame

[01:23] First, let's think of an extreme case where

[01:27] Instead of storing each one of these identical

[01:32] the first frame and repeat it N times.

[01:34] That would save us a lot of space.

[01:36] Now let's think of a more realistic case where

[01:41] This time we can do the same thing, but more

[01:46] and repeating only the blocks that don't change.

[01:49] What if all blocks change between consecutive

[01:53] a little?

[01:54] Instead of checking whether a block has changed

[01:58] the next frame within a neighborhood.

[02:01] This process is called block motion estimation.

[02:03] How does this help with compression?

[02:05] Well, instead of saving every frame, we can

[02:10] for the blocks.

[02:11] The motion vectors tell us how we should move

[02:15] This is called motion compensation.

[02:18] Although motion compensation can reduce the

[02:21] greatly, it is usually not enough by itself

[02:26] So, in addition to the motion vectors, we

[02:30] the actual and motion compensated frames.

[02:33] These differences are known as residual frames.

[02:36] When it's time to play this video, the decoder

[02:40] reference frame, compensating for the motion

[02:45] frame.

[02:46] You might ask couldn't we just save the original

[02:51] We could, but residual frames have much less

[02:56] Therefore they are highly compressible.

[02:58] Let's overview the entire process.

[03:01] Traditional video compression algorithms represent

[03:05] followed by residual frames.

[03:06] There are two types of compression done here:

[03:11] In this video, I focused mostly on the inter-frame

[03:16] efficiency by exploiting the similarities

[03:21] Intra-frame coding, on the other hand, compresses

[03:25] information within the frame and storing the

[03:30] Check out my previous video to learn more

[03:33] The methods I covered here are the very basics

[03:37] mainstream H.264 codec, which is also known

[03:43] Modern video codecs, including H.264, H.265,

[03:49] the level of compression and perceptual image

[03:54] complexity.

[03:56] Although video compression algorithms we use

[04:00] is still an active area of research.

[04:03] Researchers have already been experimenting

[04:06] potential to perform better than today's block-based

[04:11] It's not easy to beat today's encoding standards

[04:16] tuned in many possible ways.

[04:18] I still think that an end to end trainable

[04:22] compression methods, by optimizing perceptual

[04:29] That's all for today!

[04:30] I hope you liked it.

[04:31] If you have any comments or questions, let

[04:34] Subscribe for more videos.

[04:35] As always, thanks for watching, stay tuned,