哈夫曼编码的实现c语言,哈夫曼树算法c语言
这篇文章主要为大家详细介绍了如何利用计算机编程语言和C语言分别实现哈夫曼编码,文中的示例代码讲解详细,感兴趣的小伙伴可以跟随小编一起学习一下
目录
1.C语言实现1.1代码说明1.2运行结果2.计算机编程语言实现2.1代码说明2.2运行结果
1.C语言实现
1.1代码说明
a创建双向链表:
在创建哈夫曼树的过程中,需要不断对结点进行更改和删除,所以选用双向链表的结构更容易
C
#包含标准库
#包含标准视频
#包含windows.h
//哈夫曼树结构体,数据域存储字符及其权重
数据类型说明结构节点
{
char c;
(同Internationalorganizations)国际组织重量;
结构节点*lchild,* rchild
}霍夫曼,*树;
//双向链表结构体,数据域存储哈夫曼树结点
数据类型说明结构列表
{
树根;
结构列表*之前
结构列表*下一个
}列表,* pList
//创建双向链表,返回头结点指针
pList creatList()
{
pList head=(pList)malloc(sizeof(List));
pList temp1=头
pList temp 2=(pList)malloc(sizeof(List));
temp 1-pre=NULL;
temp 1-next=temp 2;
temp1-root=(树)malloc(sizeof(霍夫曼));
temp 1-root-c= a ;
温度1-根-重量=22;
temp 1-root-l child=NULL;
temp 1-root-rchild=NULL;
温度2-pre=温度1;
温度1=温度2
temp 2=(pList)malloc(sizeof(List));
temp 1-next=temp 2;
temp1-root=(树)malloc(sizeof(霍夫曼));
temp 1-root-c= b ;
温度1-根-重量=5;
temp 1-root-l child=NULL;
temp 1-root-rchild=NULL;
温度2-pre=温度1;
温度1=温度2
temp 2=(pList)malloc(sizeof(List));
temp 1-next=temp 2;
temp1-root=(Tree)malloc(sizeof(霍夫曼));
temp 1-root-c= c ;
温度1-根-重量=38;
temp 1-root-l child=NULL;
temp 1-root-rchild=NULL;
温度2-pre=温度1;
温度1=温度2
temp 2=(pList)malloc(sizeof(List));
temp 1-next=temp 2;
temp1-root=(树)malloc(sizeof(霍夫曼));
temp 1-root-c= d ;
温度1-根-重量=9;
temp 1-root-l child=NULL;
temp 1-root-rchild=NULL;
温度2-p
re = temp1;
temp1 = temp2;
temp2 = (pList)malloc(sizeof(List));
temp1->next = temp2;
temp1->root = (Tree)malloc(sizeof(Huffman));
temp1->root->c = e;
temp1->root->weight = 44;
temp1->root->lchild = NULL;
temp1->root->rchild = NULL;
temp2->pre = temp1;
temp1 = temp2;
temp2 = (pList)malloc(sizeof(List));
temp1->next = temp2;
temp1->root = (Tree)malloc(sizeof(Huffman));
temp1->root->c = f;
temp1->root->weight = 12;
temp1->root->lchild = NULL;
temp1->root->rchild = NULL;
temp2->pre = temp1;
temp1 = temp2;
temp1->next = NULL;
temp1->root = (Tree)malloc(sizeof(Huffman));
temp1->root->c = g;
temp1->root->weight = 65;
temp1->root->lchild = NULL;
temp1->root->rchild = NULL;
return head;
}
b创建栈结构:
解码过程需要用到两个栈,一个用来存放树结点,一个用来存放码0和1
C#define STACK_INIT_SIZE 100 //栈初始开辟空间大小
#define STACK_INCREMENT 10 //栈追加空间大小
//字符栈结构体,存放编码0和1
typedef struct {
char *base;
char *top;
int size;
}charStack;
//栈初始化
charStack charStackInit()
{
charStack s;
s.base = (char *)malloc(sizeof(char)*STACK_INIT_SIZE);
s.top = s.base;
s.size = STACK_INIT_SIZE;
return s;
}
//入栈
void charPush(charStack *s, char e)
{
if(s->top - s->base >= s->size)
{
s->size += STACK_INCREMENT;
s->base = realloc(s->base, sizeof(char)*s->size);
}
*s->top = e;
s->top++;
}
//出栈
char charPop(charStack *s)
{
if(s->top != s->base)
{
s->top--;
return *s->top;
}
return -1;
}
//得到栈顶元素,但不出栈
char charGetTop(charStack *s)
{
s->top--;
char temp = *s->top;
s->top++;
return temp;
}
//栈结构体,存放哈夫曼树结点
typedef struct
{
Huffman *base;
Huffman *top;
int size;
}BiStack;
//栈初始化
BiStack stackInit()
{
BiStack s;
s.base = (Huffman *)malloc(sizeof(Huffman)*STACK_INIT_SIZE);
s.top = s.base;
s.size =STACK_INIT_SIZE;
return s;
}
//入栈
void push(BiStack *s, Huffman e)
{
if(s->top - s->base >= s->size)
{
s->size += STACK_INCREMENT;
s->base = (Huffman *)realloc(s->base, sizeof(Huffman)*s->size);
}
*s->top = e;
s->top++;
}
//出栈
Huffman pop(BiStack *s)
{
Huffman temp;
s->top--;
temp = *s->top;
return temp;
}
//得到栈顶元素,但不出栈
Huffman getTop(BiStack s)
{
Huffman temp;
s.top--;
temp = *s.top;
return temp;
}
char stack[7][10]; //记录a~g的编码
//遍历栈,得到字符c的编码
void traverseStack(charStack s, char c)
{
int index = c - a;
int i = 0;
while(s.base != s.top)
{
stack[index][i] = *s.base;
i++;
s.base++;
}
}
c 创建哈夫曼树:
C//通过双向链表创建哈夫曼树,返回根结点指针
Tree creatHuffman(pList head)
{
pList list1 = NULL;
pList list2 = NULL;
pList index = NULL;
Tree root = NULL;
while(head->next != NULL) //链表只剩一个结点时循环结束,此结点数据域即为哈夫曼树的根结点
{
list1 = head;
list2 = head->next;
index = list2->next;
root = (Tree)malloc(sizeof(Huffman));
while(index != NULL) //找到链表中权重最小的两个结点list1,list2
{
if(list1->root->weight > index->root->weight list2->root->weight > index->root->weight)
{
if(list1->root->weight > list2->root->weight) list1 = index;
else list2 = index;
}
index = index->next;
}
//list1和list2设为新结点的左右孩子
if(list2->root->weight > list1->root->weight)
{
root->lchild = list1->root;
root->rchild = list2->root;
}
else
{
root->lchild = list2->root;
root->rchild = list1->root;
}
//新结点字符统一设为空格,权重设为list1与list2权重之和
root->c = ;
root->weight = list1->root->weight + list2->root->weight;
//list1数据域替换成新结点,并删除list2
list1->root = root;
list2->pre->next = list2->next;
if(list2->next != NULL)
list2->next->pre = list2->pre;
}
return head->root;
}
d编码:
Cchar stack[7][10]; //记录a~g的编码
//遍历栈,得到字符c的编码
void traverseStack(charStack s, char c)
{
int index = c - a;
int i = 0;
while(s.base != s.top)
{
stack[index][i] = *s.base;
i++;
s.base++;
}
}
//通过哈夫曼树编码
void encodeHuffman(Tree T)
{
BiStack bs = stackInit();
charStack cs = charStackInit();
Huffman root = *T;
Tree temp = NULL;
push(&bs, root); //根结点入栈
while(bs.top != bs.base) //栈空表示遍历结束
{
root = getTop(bs);
temp = root.lchild; //先访问左孩子
while(temp != NULL) //左孩子不为空
{
//将结点左孩子设为空,代表已访问其左孩子
root.lchild = NULL;
pop(&bs);
push(&bs, root);
//左孩子入栈
root = *temp;
temp = root.lchild;
push(&bs, root);
//0入字符栈
charPush(&cs, 0);
}
temp = root.rchild; //后访问右孩子
while(temp == NULL) //右孩子为空,代表左右孩子均已访问,结点可以出栈
{
//结点出栈
root = pop(&bs);
//寻到叶子结点,可以得到结点中字符的编码
if(root.c != )
traverseStack(cs, root.c);
charPop(&cs); //字符栈出栈
if(bs.top == bs.base) break; //根结点出栈,遍历结束
//查看上一级结点是否访问完左右孩子
root = getTop(bs);
temp = root.rchild;
}
if(bs.top != bs.base)
{
//将结点右孩子设为空,代表已访问其右孩子
root.rchild = NULL;
pop(&bs);
push(&bs, root);
//右孩子入栈
root = *temp;
push(&bs, root);
//1入字符栈
charPush(&cs, 1);
}
}
}
e解码:
Cchar decode[100]; //记录解码得到的字符串
//通过哈夫曼树解码
void decodeHuffman(Tree T, char *code)
{
int cnt = 0;
Tree root;
while(*code != \0) //01编码字符串读完,解码结束
{
root = T;
while(root->lchild != NULL) //找到叶子结点
{
if(*code != \0)
{
if(*code == 0)
root = root->lchild;
else
root = root->rchild;
code++;
}
else break;
}
decode[cnt] = root->c; //叶子结点存放的字符即为解码得到的字符
cnt++;
}
}
f主函数:
Cvoid main()
{
pList pl = creatList();
printf("字符的权重如下\n");
for(pList l = pl; l->next != NULL; l = l->next)
printf("字符%c的权重是 %d\n", l->root->c, l->root->weight);
Tree T = creatHuffman(pl);
encodeHuffman(T);
printf("\n\n字符编码结果如下\n");
for(int i = 0; i < 7; i++)
printf("%c : %s\n", i+a, stack[i]);
char code[100];
printf("\n\n请输入编码:\n");
scanf("%s", code);
printf("解码结果如下:\n");
decodeHuffman(T, code);
printf("%s\n", decode);
printf("\n\n");
system("date /T");
system("TIME /T");
system("pause");
exit(0);
}
1.2运行结果
2.Python实现
2.1代码说明
a创建哈夫曼树:
#coding=gbkimport datetime
import time
from pip._vendor.distlib.compat import raw_input
#哈夫曼树结点类
class Huffman:
def __init__(self, c, weight):
self.c = c
self.weight = weight
self.lchild = None
self.rchild = None
#创建结点左右孩子
def creat(self, lchild, rchild):
self.lchild = lchild
self.rchild = rchild
#创建列表
def creatList():
list = []
list.append(Huffman(a, 22))
list.append(Huffman(b, 5))
list.append(Huffman(c, 38))
list.append(Huffman(d, 9))
list.append(Huffman(e, 44))
list.append(Huffman(f, 12))
list.append(Huffman(g, 65))
return list
#通过列表创建哈夫曼树,返回树的根结点
def creatHuffman(list):
while len(list) > 1: #列表只剩一个结点时循环结束,此结点即为哈夫曼树的根结点
i = 0
j = 1
k = 2
while k < len(list): #找到列表中权重最小的两个结点list1,list2
if list[i].weight > list[k].weight or list[j].weight > list[k].weight:
if list[i].weight > list[j].weight:
i = k
else:
j = k
k += 1
root = Huffman( , list[i].weight + list[j].weight) #新结点字符统一设为空格,权重设为list1与list2权重之和
if list[i].weight < list[j].weight: #list1和list2设为新结点的左右孩子
root.creat(list[i], list[j])
else:
root.creat(list[j], list[i])
#list1数据域替换成新结点,并删除list2
list[i] = root
list.remove(list[j])
return list[0]
b编码:
#通过哈夫曼树编码def encodeHuffman(T):
code = [[], [], [], [], [], [], []]
#列表实现栈结构
treeStack = []
codeStack = []
treeStack.append(T)
while treeStack != []: #栈空代表遍历结束
root = treeStack[-1]
temp = root.lchild
while temp != None:
#将结点左孩子设为空,代表已访问其左孩子
root.lchild = None
#左孩子入栈
treeStack.append(temp)
root = temp
temp = root.lchild
#0入编码栈
codeStack.append(0)
temp = root.rchild #后访问右孩子
while temp == None: #右孩子为空,代表左右孩子均已访问,结点可以出栈
root = treeStack.pop() #结点出栈
#寻到叶子结点,可以得到结点中字符的编码
if root.c != :
codeTemp = codeStack.copy()
code[ord(root.c) - 97] = codeTemp
if treeStack == []: #根结点出栈,遍历结束
break
codeStack.pop() #编码栈出栈
#查看上一级结点是否访问完左右孩子
root = treeStack[-1]
temp = root.rchild
if treeStack != []:
treeStack.append(temp) #右孩子入栈
root.rchild = None #将结点右孩子设为空,代表已访问其右孩子
codeStack.append(1) #1入编码栈
return code
c解码:
#通过哈夫曼树解码def decodeHuffman(T, strCode):
decode = []
index = 0
while index < len(strCode): #01编码字符串读完,解码结束
root = T
while root.lchild != None: #找到叶子结点
if index < len(strCode):
if strCode[index] == 0:
root = root.lchild
else:
root = root.rchild
index += 1
else:
break
decode.append(root.c) #叶子结点存放的字符即为解码得到的字符
return decode
d主函数:
if __name__ == __main__:list = creatList()
print("字符的权重如下")
for i in range(len(list)):
print("字符{}的权重为: {}".format(chr(i+97), list[i].weight))
T = creatHuffman(list)
code = encodeHuffman(T)
print("\n字符编码结果如下")
for i in range(len(code)):
print(chr(i+97), end= : )
for j in range(len(code[i])):
print(code[i][j], end=)
print("")
strCode = input("\n请输入编码:\n")
#哈夫曼树在编码时被破坏,必须重建哈夫曼树
list = creatList()
T = creatHuffman(list)
decode = decodeHuffman(T, strCode)
print("解码结果如下:")
for i in range(len(decode)):
print(decode[i], end=)
print("\n\n")
datetime = datetime.datetime.now()
print(datetime.strftime("%Y-%m-%d\n%H:%M:%S"))
input("Press Enter to exit…")
2.2运行结果
以上就是利用Python和C语言分别实现哈夫曼编码的详细内容,更多关于Python哈夫曼编码的资料请关注盛行IT软件开发工作室其它相关文章!
郑重声明:本文由网友发布,不代表盛行IT的观点,版权归原作者所有,仅为传播更多信息之目的,如有侵权请联系,我们将第一时间修改或删除,多谢。