1 A quick description of Rate distortion theory.
3 We want to encode a video, picture or music optimally.
4 What does optimally mean?
5 It means that we want to get the best quality at a given
6 filesize OR (which is almost the same actually) We want to get the
7 smallest filesize at a given quality.
9 Solving this directly isnt practical, try all byte sequences
10 1MB long and pick the best looking, yeah 256^1000000 cases to try ;)
12 But first a word about Quality also called distortion, this can
13 really be almost any quality meassurement one wants. Commonly the
14 sum of squared differenes is used but more complex things that
15 consider psychivisual effects can be used as well, it makes no differnce
19 First step, that RD factor called lambda ...
20 Lets consider the problem of minimizing
22 distortion + lambda*rate
24 for a fixed lambda, rate here would be the filesize, distortion the quality
25 Is this equivalent to finding the best quality for a given max filesize?
26 The awnser is yes, for each filesize limit there is some lambda factor for
27 which minimizing above will get you the best quality (in your provided quality
28 meassurement) at that (or a lower) filesize
31 Second step, spliting the problem.
32 Directly spliting the problem of finding the best quality at a given filesize
33 is hard because we dont know how much filesize to assign to each of the
34 subproblems optimally.
35 But distortion + lambda*rate can trivially be split
37 (distortion0 + distortion1) + lambda*(rate0 +rate1)
38 a problem made of 2 independant subproblems, the subproblems might be 2
39 16x16 macroblocks in a frame of 32x16 size.
41 (distortion0 + distortion1) + lambda*(rate0 +rate1)
42 one just have to minimize
43 distortion0 + lambda*rate0
45 distortion1 + lambda*rate1
47 aka the 2 problems can be solved independantly
49 Author: Michael Niedermayer