Computer Networks · 免费预览
多媒体数据与压缩
多媒体数据若不压缩就会淹没任何网络，因此端到端数据工程在很大程度上就是压缩的艺术：无损编码（如「Huffman」编码）将较短的比特串分配给大概率符号，行程编码则压缩连续的重复序列；图像方面，图形使用调色板索引的无损GIF，而JPEG采用变换编码——它将图像分块，做离散余弦变换，量化掉人眼看不见的高频成分，再对剩余部分进行熵编码；视频则增加了时间维度，MPEG交替使用自包含的I帧、由过去预测的P帧，以及通过运动矢量从两个方向插值的B帧；音频部分以心理声学原理收尾，MP3丢弃那些掩蔽效应表明无人会听到的声音。
1. Squeezing Without Losing
Squeezing Without Losing
2. Transforming Away What Eyes Miss
Transforming Away What Eyes Miss
3. Borrowing From the Timeline
Borrowing From the Timeline
4. Hearing What Isn't There
Hearing What Isn't There
5. The entropy bargain
What is the essential idea behind Huffman coding?
6. Choosing the image format
A crisp company logo with eight flat colors needs a web format. Why does GIF fit better than JPEG?
7. The lossy step
In JPEG's pipeline, which step actually discards information?
8. Frames that lean
What distinguishes MPEG's B frames from P frames?
9. The cascade
Why does losing one I frame damage more playback than losing one B frame?
10. The psychoacoustic principle
On what principle does MP3 decide what not to encode?
11. The repeated-run trick
What lossless technique replaces a run of repeated symbols with one copy of the symbol and a count?
12. The 8x8 change of basis
Which transform does JPEG apply to each 8x8 block to re-express pixels as spatial frequency coefficients?
13. The moving reference
In MPEG prediction, what does a macroblock's motion vector describe?
14. The hidden sounds
What psychoacoustic phenomenon lets MP3 omit sounds near loud tones?