Types of compression algorithms

http://www.html5rocks.com/en/tutorials/speed/img-compression/

Types of compression algorithms

There are generally two stages in an image compressor, a lossy phase, and lossless phase. Lossy compression algorithms will modify the source stream such that you lose information that cannot be restored upon decompression.Most lossy algorithms in image compression take advantage of how the human visual system works, often removing information that we really can’t see, and in the process, saving bytes. For example, limiting the colors used in an image; fewer colors means there’s less data to run around. Generally, when you save an image in a format supporting Lossy compression, you’re asked what “quality level” you’d like for the image, effectively, what you’re choosing is a scalar value which trades file-size for image-quality. Savvy web developers realize that there is a sweet-spot for images, such that the quality level is high enough, and the file size is low as possible.

Before	After
0.123, 1.2345, 21.2165, 21.999, 12.123	0,0,20,20,10

Figure 1 - An example of lossy compression. Values are quantized to the smallest multiple of 10 they occupy. This transform cannot be reversed.

After a lossy compressor, a lossless variant is then applied, that is, the data, once uncompressed, is restored to it’s exact state, before compression. These are typical compression algorithms that allow the source stream to be recovered directly without any loss of precision or information. In Images, popular Lossless codecs include LZ77, RLE, and Arithmetic encoding. Lossless compression algorithms are the backbone of compression, often squeezing out the last percentages of data from your content, constantly struggling with information theory to reduce your data sizes.

Before	After
aaaaabbbbbcccddddeeeeffffaaaaabb	a5b4c2d4e4f4a5bb0

Figure 2 - An example of lossless compression. Runs of values are encoded as the symbol followed by the length of the run. We can properly restore the origional stream. Note that if the length of the run is <= 2 characters, it makes sense to just leave the symbols alone. You see this at the end of the stream with ‘bb’.

小结：

0-可还原性，可逆性。

发问：

0-从0101层面考虑呢？