算术编码介绍
算术压缩是一种无损的熵编码。1948年,香农提出将信源符号依其概率降序排序,用符号序列累计的二进制作为对信源数据的编码。
-
- 1960年,Peter Elias发现无需排序,只要编解码端使用相同的符号顺序即可,提出了算术编码的概念。
-
- 1967年,R. Pasco和J. Rissanen分别用丁昌德寄存器实现了有限精度的算术编码。
-
- 1979年,Rissanen和G. G. Langdon一起将算术编码系统化,并于1981年实现了二进制编码。
-
- 1987年,Witten等人发表了一个实用的算术编码程序,即CACM87,后用于ITUT的H.263视频压缩标准。同期,IMB公司发表了著名的Q-编码器,后用于JPEG和JBIG图像压缩标准。
传统的算术压缩—编码
假设信源符号有5个,分别是a、b、c、d、e这5种,其中出现的概率为:
假如有以下待压缩的数据“abcade”,以区间[0,1]为编码范围。
-
- 第一个字符为a,将[0,1]等分为5份[0,0.2,0.4,0.6,0.8,1],a在[0,0.2]中;
-
- 第二个字符为b ,取1步中的[0,0.2]等分为5份[0,0.04,0.08,0.12,0.16,0.2],b在[0.04,0.08]中;
-
- 第三个字符为c ,取2步中的[0.04,0.08]继续等分为5份,分别为[0.04,0.048,0.056,0.064,0.072,0.08],c在[0.056,0.064]中;
-
- 第四个字符为a ,取3步中的[0.056,0.064]继续等分为5份,分别为[0.056,0.0576, 0.0592,0.0608, 0.0624, 0.064],a在[0.056,0.0576]中;
-
- 第五个字符为d ,取4步中的[0.056,0.0576]继续等分为5份,分别为[0.056,0.05632, 0.05664,0.05696, 0.05728, 0.0576],d在[0.05696, 0.05728]中;
-
- 第六个字符为e ,取5步中的[0.05696, 0.05728]继续等分为5份,分别为[0.05696,0.056992, 0.057024,0.057088, 0.057216, 0.05728],e在[0.057216, 0.05728]中;
-
- 最终“abcade”编码的区间为[0.057216, 0.05728],在其中任取一个值即可代表此数据。
-
- 将区间[0.057216, 0.05728]分别转为二进制编码为00001110101001011011010100110000110011101101010011100101、000011101010100111100110111011101011011100000010011
-
- 取值取到上区间和下区间相对位置相同的后一位,上面最终编码为:0000111010101 转为十进制为 0.0572509765625,正好在最终区间范围内
最终,abcade被压缩为0000111010101。
压缩前 abcade所占8*6=48比特,而算术压缩后所占13比特,压缩率达到13/48=27%
- 取值取到上区间和下区间相对位置相同的后一位,上面最终编码为:0000111010101 转为十进制为 0.0572509765625,正好在最终区间范围内
传统的算术压缩—解码
假设信源符号有5个,分别是a、b、c、d、e这5种,其中出现的概率为:
假如有以下压缩后的数据“0000111010101”,以区间[0,1]为编码范围。
-
- 0000111010101 转10进制为 0.0572509765625;
-
- [0,1]等分为5份[0,0.2,0.4,0.6,0.8,1],0.0572509765625在[0,0.2]中 ,第一个字符为a;
-
- [0,0.2]等分为5份[0,0.04,0.08,0.12,0.16,0.2],0.0572509765625在[0.04,0.08]中,第二个字符为b;
-
- [0.04,0.08]等分为5份[0.04,0.048,0.056,0.064,0.072,0.08],0.0572509765625在[0.056,0.064]中,第三个字符为c;
-
- [0.056,0.064]等分为5份[0.056,0.0576, 0.0592,0.0608, 0.0624, 0.064],0.0572509765625在[0.056,0.0576]中,第四个字符为a;
-
- [0.056,0.0576]等分为5份[0.056,0.05632, 0.05664,0.05696, 0.05728, 0.0576],0.0572509765625在[0.05696, 0.05728]中,第五个字符为d;
-
- [0.05696, 0.05728]等分为5份[0.05696,0.056992, 0.057024,0.057088, 0.057216, 0.05728],0.0572509765625在[0.057216, 0.0572]中,第六个字符为e;
-
- 最终还原编码“abcade”。
传统的算术压缩—伪代码
传统的算术压缩—伪代码–编码
set Low to 0
set High to 1
while there are input symbols do
take a symbol
CodeRange = High – Low
High = Low + CodeRange *HighRange(symbol)
Low = Low + CodeRange * LowRange(symbol)
end of while
output Low
传统的算术压缩—伪代码–解码
get encoded number
do
find symbol whose range straddles the encoded number
output the symbol
range = symbo.LowValue – symbol.HighValue
substract symbol.LowValue from encoded number
divide encoded number by range
until no more symbols