python 实现的huffman 编码压缩，解码解压缩
刚刚实现一个初始版本1.TODO 仅仅能处理英文，下一步考虑unicode
似乎考虑多了，当前的程序处理中文文本是一样可以的。
2.TODO enocde ,decode，文本读写多重转换 int -> chrchr -> int -> bin
下一步直接读写int，能否直接读写bit?
3.TODO 其它方面考虑速度的优化，比如垃圾回收机制是否影响了速度等等，
和c/c++比python肯定没有速度优势，不过代码写起来比c/c++舒服多了，感觉python非常接近写伪码的感觉了，所想即所得，
一个问题只有一个解法，真正让你能够专注与算法与框架流程设计而不被语言本身所束缚。
5.TODO 设计成可以对一些文本一起压缩，全部解压缩或者对指定文本解压缩。
5.特别利用pygraphivz对huffman tree 进行了绘制，有利于调试。见前一篇随笔。
6.TODO 考虑其它压缩方法，如范式huffman 的实现。分词后对词编码而不是对字母编码的方法。
7.压缩过程中写文件遇到的一个问题，因为我只有到扫描处理完所有文件的字符a,b,c...才能计算出最后一个字节剩余了多少个bit，它们被补0，而计算好之后我希望把这个信息写到前面，即压缩文档开头序列化之后马上先记录最后一个byte补了多少个0，然后记录最后一个byte,然后从头到尾顺序记录所有其它被encode translate的byte，所以我先保持了原来的需要写的位置，当写到最后的时候，再把写指针指回，写那个位置，但是我在解压缩过程再读那个位置的时候发现最后的写操作并没有写成功。
self.infile.seek(0)
#save this pos we will write here later
pos = self.outfile.tell()
self.outfile.write(chr(0))#store left bit
self.outfile.write(chr(0))#if left bit !=0 this is the last byte
#.... translate other bytes
#just after the huffman tree sotre how many bits are left for last
#byte that is not used and filled with 0
self.outfile.seek(pos)
self.outfile.write(chr(leftBit))#still wrong can't not read well
self.outfile.write(chr(num))
后来发现要再最后加一句self.outfile.flush()将内容写回硬盘，问题似乎是回写前面的位置，仅仅写到了cache中，最后file.close()的时候该cache的新内容也未被写回硬盘，不知道是不是python2.6的一个bug
反正最后加file.flush()就ok了。
当前流程
压缩过程：
读文本
计算各个字符的出现频率
建立huffman tree （二叉连表树实现，不需要parent域)
通过huffman tree 为每个字符编码（深度优先遍历huffman tree，即可得到所以字符编码)
将huffman tree 序列化写到输出文本（以便解压缩时恢复huffman tree，这里采用存储huffman tree 的前序序列，根据huffman tree的特性，每个内部节点均2度，可恢复）
再读文本，为每个字符通过dict 取出它的编码并且写到输出文本。
（注意写的时候集齐8个字符为一组，输出，但是最后一个byte可能不够8位，要用0补齐空位。
为了处理方便，我将在序列化的二叉树后面首先记录最后一个byte需要用0补齐的位数，如果需要补齐的位 数不为0，则接下来输出最后一个byte，然后再从输入文件内部头开始
编码输出到输出文件。这里的技巧就是把最后一个byte放到了前面，便于处理，否则解码可能最后文件尾部 会有多余字符被译出。）
解压缩过程：
读压缩好的文本
先读文件头部，根据huffman tree前序序列，恢复建立huffman tree，二叉链表树
继续读文本，根据huffman tree 进行解码，0向左，1向右，到叶节点，解码一个字符。
解码输出完成即完成解压缩。（注意我压缩的时候最后一个byte放到前面了，如果需要要
将其最后输出。）
当前程序用法
python2.6 huffman.pyinput.txt
输出
input.txt.compress压缩文件
input.txt.compress.de解压缩后的，内容应与input.txt一致。
allen:~/study/data_structure/huffman$ time python2.6 huffman.py C00-1052.txt
real0m0.607s
user0m0.536s
sys0m0.060s
allen:~/study/data_structure/huffman$ diff C00-1052.txt C00-1052.txt.compress.de
allen:~/study/data_structure/huffman$ du -h C00-1052.txt
36KC00-1052.txt
allen:~/study/data_structure/huffman$ du -h C00-1052.txt.compress.de
36KC00-1052.txt.compress.de
allen:~/study/data_structure/huffman$ du -h C00-1052.txt.compress
24KC00-1052.txt.compress
网上有不少关于huffman的实现，和我这里一样都是采用最简单的基本huffman算法。
做了下对比，采用《平凡的世界》1.7M, 似乎python的效率还不错，不过应该用更大
的文件对比下。另外为什么http://www.javaresearch.org/article/97725.htm中的实现
的压缩比率更大呢，应该压缩率一样的啊。
allen:~/study/data_structure/huffman$ time python2.6 huffman.py normal_world.log
real0m32.236s
user0m31.298s
sys0m0.732s
allen:~/study/data_structure/huffman$ du -h normal_world.log
1.7Mnormal_world.log
allen:~/study/data_structure/huffman$ du -h normal_world.log.compress
1.3Mnormal_world.log.compress
allen:~/study/data_structure/huffman$ du -h normal_world.log.compress.de
1.7Mnormal_world.log.compress.de
allen:~/study/data_structure/huffman$ diff normal_world.log normal_world.log.compress.de
原文件《平凡的世界》，大小1.7M，压缩后1.3M，解压缩后与原文件完全相同，压缩和解压缩共耗时32s
对比http://www.javaresearch.org/article/97725.htm,该java版本，作者提到
压缩效果
使用本程序对《平凡的世界》做压缩测试，压缩前为文本文件，大小为1.7M,压缩后为二进制文件，大小接近1M（988,817byte）,而zip压缩后体积为920,997byte,比zip差，压缩文件存储格式待改善。另外，因为从Huffman压缩算法的原理可知，该算法对字符重复率高的文本最有效，比如长篇小说或者英文小说。
另外网上有一个c版本的huffman，http://blog.sina.com.cn/s/blog_4ab057eb0100bx34.html
作者提到：
l略大文件
test3.txt 《平凡的世界》
压缩前：1.62M
压缩后：1.39M
压缩率：86%
压缩时间14.23秒
解压时间 16.85秒
测试结果：压缩，解压成功！
压缩解压时间在可接受范围之内
1 '''
2 Create a huffman tree from
3 the input is a list like
4 [('a',3), ('b',2)]
5 frequnce of 'a' appeard is stored as it's weight
6 '''
7 from Queue import PriorityQueue
8 #if do not use treeWiter so not include pygraphviz than can use py3.0
9 from treeWriter import TreeWriter
10 from copy import copy
11
12 class NodeBase():
13 def __init__(self):
14 self.weight = 0
15
16 def elem(self):
17 return self.weight
18
19 class Node(NodeBase):
20 def __init__(self, weight = 0, left = None, right = None):
21 self.weight = weight
22 self.left = left
23 self.right = right
24
25 def __str__(self):
26 return str(self.weight)
27
28 class Leaf(NodeBase):
29 def __init__(self, key = '', weight = 0):
30 self.key = key
31 self.weight = weight
32
33 def __str__(self):
34 return str(self.key)
35
36
37 def convert(c):
38 '''
39 input c = 'a' ord(a) = 97
40 bin(97) = '0b1100001'
41 return ['0', '1', '1', '0', '0', '0', '0', '1']
42 '''
43 l1 = list(bin(ord(c))) #like 0b11101
44 l2 = ['0'] * (10 - len(l1))
45 l2.extend(l1)
46 return l2
47
48 class HuffmanTree():
49 '''
50 base class for HuffmanTreeForCompress and HuffmanTreeForDecompress
51 '''
52 def __init__(self):
53 self.root = None
54
55 class HuffmanTreeForCompress(HuffmanTree):
56 '''
57 create a huffman tree for the compressing process
58 here self.list like [('a',3),('b',4)] where 'a' is key, 3 is weight
59 or say frequence of 'a' appear in the text
60 '''
61 def __init__(self, list):
62 HuffmanTree.__init__(self)
63 self.list = list #like [('a',3),('b',4)]
64 self.dict = {} #like {'a': , .}
65
66 self.__buildTree()
67 self.__genEncode()
68
69 def __initPriorityQueue(self, queue):
70 '''
71 init priority queue let lowest weight at top
72 '''
73 for key, weight in self.list:
74 leaf = Leaf(key, weight)
75 queue.put((weight,leaf))
76
77 def __buildTree(self):
78 '''
79 build the huffman tree from the list of weight using prority queue
80 greedy alogrithm,choose two least frequence node first
81 '''
82 length = len(self.list)
83 queue = PriorityQueue(length)
84 self.__initPriorityQueue(queue)
85 #while queue.qsize() > 1:
86 # do len(self.list) - 1 times same as while queue.qsize() > 1
87 for i in range(length - 1):
88 left = queue.get()
89 right = queue.get()
90 weight = left.weight + right.weight
91 node = Node(weight, left, right)
92 queue.put((weight,node))
93 self.root = queue.get()
94
95 def __genEncode(self):
96 '''
97 get huffman encode for each key using depth first travel of tree
98 '''
99 def genEncodeHelp(root, encode = []):
100 if isinstance(root, Leaf):
101 #TODO notice need copy content here,why can't list(encode)?
102 self.dict = copy(encode)
103 #print self.dict
104 return
105 encode.append(0)
106 genEncodeHelp(root.left, encode)
107 encode = 1
108 genEncodeHelp(root.right, encode)
109 encode.pop()
110 genEncodeHelp(self.root)
111
112
113 class HuffmanTreeForDecompress(HuffmanTree):
114 '''
115 rebuild of huffman tree for the decompressing process
116 '''
117 def __init__(self, infile):
118 HuffmanTree.__init__(self)
119 self.__buildTree(infile)
120
121 def __buildTree(self, infile):
122 def buildTreeHelp(infile):
123 first = infile.read(1)
124 second = infile.read(1)
125 #if not (first == '\xff' and second == '\xfe'):#is leaf
126 if first == '\x00':#is leaf, not consider unicode now
127 return Leaf(second)
128 node = Node()
129 node.left = buildTreeHelp(infile)
130 node.right = buildTreeHelp(infile)
131 return node
132 infile.read(2)
133 self.root = Node()
134 self.root.left = buildTreeHelp(infile)
135 self.root.right = buildTreeHelp(infile)
136
137 class Decompress():
138 def __init__(self, infileName, outfileName = ''):
139 #TODO better name, expection of opening file
140 self.infile = open(infileName, 'rb')
141 if outfileName == '':
142 outfileName = infileName + '.de'
143 self.outfile = open(outfileName, 'wb')
144 self.tree = None
145
146 def __del__(self):
147 self.infile.close()
148 self.outfile.close()
149
150 def decompress(self):
151 self.__rebuildHuffmanTree()
152 self.__decodeFile()
153
154 def __rebuildHuffmanTree(self):
155 self.infile.seek(0)
156 self.tree = HuffmanTreeForDecompress(self.infile)
157 #HuffmanTreeWriter(self.tree).write('tree2.png') #for debug
158
159 def __decodeFile(self):
160 #right now do not consier speed up using table
161 #do not consider the last byte since it's wrong right now
162
163 #TODO use a table as 0x00 -> 0000 0000will speed up?
164 self.outfile.seek(0)
165 leftBit = ord(self.infile.read(1))
166 lastByte = self.infile.read(1) #it is the last byte if leftBit != 0
167 curNode = self.tree.root
168 #import gc
169 #gc.disable()
170 while 1:
171 c = self.infile.read(1) #how about Chinese caracter? 2 bytes?
172 if c == '':
173 break
174 li = convert(c) #in c++ you can not return refernce to local in func here ok? yes
175 for x in li:
176 if x == '0':
177 curNode = curNode.left
178 else:
179 curNode = curNode.right
180 if isinstance(curNode, Leaf): #the cost of isinstance is higer than lkie root.left == None ?
181 self.outfile.write(curNode.key)
182 curNode = self.tree.root
183
184
185 #deal with the last bye if leftBit != 0
186 #TODO notcice code repeate can we improve?
187 if leftBit:
188 li = convert(lastByte)
189 for x in li:
190 if x == '0':
191 curNode = curNode.left
192 else:
193 curNode = curNode.right
194 if isinstance(curNode, Leaf): #the cost of isinstance is higer than lkie root.left == None ?
195 self.outfile.write(curNode.key)
196 curNode = self.tree.root
197 break #for the last byte if we find one than it's over,the other bits are useless
198
199 self.outfile.flush()
200 #gc.enable()
201
202
203
204 class Compress():
205 def __init__(self, infileName, outfileName = ''):
206 self.infile = open(infileName, 'rb')
207 if outfileName == '':
208 outfileName = infileName + '.compress'
209 self.outfile = open(outfileName, 'wb')
210 self.dict = {}
211 self.tree = None
212
213 def __del__(self):
214 self.infile.close()
215 self.outfile.close()
216
217 def compress(self):
218 self.__caculateFrequence()
219 self.__createHuffmanTree()
220 self.__writeCompressedFile()
221
222 def __caculateFrequence(self):
223 '''
224 The first time of reading the input file and caculate each
225 character frequence store in self.dict
226 '''
227 self.infile.seek(0)
228 while 1:
229 c = self.infile.read(1) #how about Chinese caracter? 2 bytes?
230 if c == '':
231 break
232 #print c
233 if c in self.dict:
234 self.dict += 1
235 else:
236 self.dict = 0
237
238 def __createHuffmanTree(self):
239 '''
240 Build a huffman tree from self.dict.items()
241 '''
242 #TODO for py 3.0 need list(self.dict.items()) instead
243 self.tree = HuffmanTreeForCompress(list(self.dict.items()))
244 #HuffmanTreeWriter(self.tree).write('tree1.png') #for debug
245
246 def __writeCompressedFile(self):
247 '''
248 Create the compressed file
249 First write the huffman tree to the head of outfile
250 than translate the input file with encode and write the result to
251 outfile
252 '''
253 self.outfile.seek(0)
254 self.__serializeTree()
255 self.__encodeFile()
256
257 def __serializeTree(self):
258 '''
259 In order to write the tree like node node leaf node .
260 in pre order sequence to the compressed file head
261 here will return the sequence list
262 TODOreuse pre order and using decorator technic!!
263 list like [(0,0), (0,0), (1,'c')],
264 (0,0) the first 0 means internal node
265 (1,'c') the first 1 means leaf and 'c' is the key
266 '''
267 def serializeTreeHelp(root, mfile):
268 if isinstance(root, Leaf):
269 mfile.write('\x00') #0x0
270 mfile.write(root.key)
271 return
272 mfile.write('\xff') #'\xff' is one character representing 0xff
273 mfile.write('\xfe') #0xfe
274 serializeTreeHelp(root.left, mfile)
275 serializeTreeHelp(root.right, mfile)
276 serializeTreeHelp(self.tree.root, self.outfile)
277
278
279 def __encodeFile(self):
280 '''
281 The second time of reading input file
282 translate the input file with encode and write the result to outfile
283 TODO can this be improved speed up?
284 just write \xff as \b 1111 1111 ? can this be possible so do not need
285 to caculate 255 than translate to \xff and write?
286 '''
287 self.infile.seek(0)
288 #save this pos we will write here later
289 pos = self.outfile.tell()
290 self.outfile.write(chr(0))#store left bit
291 self.outfile.write(chr(0))#if left bit !=0 this is the last byte
292 num = 0
293 i = 0;
294 while 1:
295 c = self.infile.read(1) #how about Chinese caracter? 2 bytes?
296 if c == '':
297 break
298 li = self.tree.dict
299 for x in li:
300 num = (num
页:
[1]