java实现哈夫曼压缩与解压缩的方法

一哈夫曼树以及文件压缩原理:

成都创新互联公司是专业的射阳网站建设公司,射阳接单;提供成都网站设计、成都网站制作,网页设计,网站设计,建网站,PHP网站建设等专业做网站服务;采用PHP框架,可快速的进行射阳网站开发网页制作和功能扩展;专业做搜索引擎喜爱的网站,专业的做网站团队,希望更多企业前来合作!

1.哈夫曼树 :

给定N个权值作为N个叶子结点,构造一棵二叉树,若该树的带权路径长度达到最小,称这样的二叉树为最优二叉树,也称为哈夫曼树。哈夫曼树是带权路径长度最短的树,权值较大的结点离根较近(频率越高的结点离根越进)。

以 下数组为例,构建哈夫曼树

int a[] = {0,1,2,3,4,5,6,7,8}

java实现哈夫曼压缩与解压缩的方法

我们可以发现以下规律

1:9个数构成的哈夫曼树一共有17个结点,也就是可以n个数可以生产2*n-1个结点

2:数字越大的数离根节点越近,越小的数离根节点越近。

2.如何利用haffman编码实现文件压缩:

比如abc.txt文件中有以下字符aaaabbbccde,

1.进行字符统计

aaaabbbccde
 
a : 4次
b : 3次
c : 2次
d : 1次
e : 1次

2.用统计结果构建哈夫曼树

java实现哈夫曼压缩与解压缩的方法

3.用哈夫曼树生成哈夫曼编码(从根结点开始,路径左边记为0,右边记为1):

a的编码:1
b的编码:01
c的编码:000
d的编码:0011
e的编码:0010

4.哈夫曼编码代替字符,进行压缩。

源文件内容为:aaaabbbccde
将源文件用对应的哈夫曼编码(haffman code)替换,则有:11110101 01000000 00110010 (总共3个字节)

由此可见,源文件一共有11个字符,占11字节的内存,但是经过用haffman code替换之后,只占3个字节,这样就能达到压缩的目的

二主要技术点:

1.哈夫曼树算法(哈夫曼压缩的基本算法)

2.哈希算法(字符统计时候会用到,也可以直接用HashMap统计)

3.位运算(涉及到将指定位,置0或置1)

4.java文件操作,以及缓冲操作。

5.存储模式(大端存储,小端存储,能看懂文件16进制的形式)

7.设置压缩密码,解压输入密码解压(小编自己加的内容)

三实现过程:

以上述aaaabbbccde为例

1.字符统计:

public class FreqHuf {
	public static int BUFFER_SIZE = 1 << 18;
	int freq[] = new int[256];
	File file;
	int count;
	List list;
	
	FreqHuf(String pathname) throws Exception {
		list = new ArrayList<>();
		this.file = new File(pathname);
		if(!file.exists()){
			throw new Exception("文件不存在");
		}
		System.out.println("进行字符统计中");
		CensusChar();
		System.out.println("字符统计完毕");
	}
	
	public void CensusChar() throws IOException{
		int intchar;
		FileInputStream fis = new FileInputStream(file);
		System.out.println("统计中");
 
//这种统计处理方案,速度极慢,不建议使用,以下采用缓存读数据。
//		while((intchar = fis.read()) != -1){
//			freq[intchar]++;
//		}
 
		//这里采用缓存机制,一次读1 << 18个字节,大大提高效率。
		byte[] bytes = new byte[BUFFER_SIZE];
		while((intchar = fis.read(bytes))!= -1){
			for(int i = 0; i < intchar;i++){
				int temp = bytes[i]& 0xff;
				freq[temp]++;
			}
		}
		
		
		
		fis.close();
		
		for(int i = 0; i < 256; i++){
			if(freq[i] != 0){
				this.count++;
			}
		}
		
		int index = 0;
		for(int i = 0; i < 256; i++){
			if(freq[i] != 0){
				HuffmanFreq huffman = new HuffmanFreq();
				huffman.character = (char)i;
				huffman.freq = freq[i];
				list.add(index, huffman);
			}
		}
	}
}
//统计每个字符和其频率的类
public class HuffmanFreq {
	char character;
	int freq;
	
	HuffmanFreq() {
	}
	
	HuffmanFreq(int character,int freq) {
		this.character = (char)character;
		this.freq = freq;
	}
 
	char getCharacter() {
		return character;
	}
 
	void setCharacter(int character) {
		this.character = (char)character;
	}
 
	int getFreq() {
		return freq;
	}
 
	void setFreq(int freq) {
		this.freq = freq;
	}
	
	byte[] infoToByte(){
		byte[] bt = new byte[6];
		
		byte[] b1 = ByteAnd8Types.charToByte(character);
		for(int i= 0; i < b1.length;i++){
			bt[i] = b1[i];
		}
		
		byte[] b2 = ByteAnd8Types.intToBytes2(freq);
		int index = 2;
		for(int i= 0; i < b2.length;i++){
			bt[index++] = b2[i];
		}
		
		return bt;
	}
 
	@Override
	public String toString() {
		return "Huffman [character=" + character + ", freq=" + freq + "]";
	}
}

2.用统计结果构建哈夫曼树:

//treeSize为总节点数
private void creatTree(int treeSize){
		int temp;
		treeList = new ArrayList();
		for(int i = 0; i < treeSize; i++){
			HuffTreeNode node = new HuffTreeNode();
			treeList.add(i, node);
		}
		
		for(int i = 0; i < charCount; i++){
			HuffTreeNode node = treeList.get(i);
			node.freq.freq = charList.get(i).getFreq();
			node.freq.character = charList.get(i).getCharacter();
			node.left = -1;
			node.right = -1;
			node.use = 0;
		}
		
		for(int i = charCount; i < treeSize; i++){
			int index = i;
			HuffTreeNode node = treeList.get(i);
			node.use = 0;
			node.freq.character = '#';
			node.right = searchmin(index);
			node.left = searchmin(index);
			node.freq.freq = treeList.get(node.right).freq.freq + treeList.get(node.left).freq.freq;
			temp = searchmin(++index);
			if(temp == -1){
				break;
			}
			treeList.get(temp).use = 0;
		}
	}
	
	private int searchmin(int count){
		int minindex = -1;
		
		for(int i = 0; i < count; i++){
			if(treeList.get(i).use == 0){
				minindex = i;
				break;
			}
		}
		if(minindex == -1){
			return -1;
		}
		for(int i = 0; i < count; i++){
			if((treeList.get(i).freq.freq <= treeList.get(minindex).freq.freq) && treeList.get(i).use == 0){
				minindex = i;
			}
		}
		treeList.get(minindex).use = 1;
		return minindex;
	}

3.用哈夫曼树生成哈夫曼编码(从根结点开始,路径左边记为0,右边记为1):

  private void bulidhuftreecode(int root, String str){
		if(treeList.get(root).getLeft() != -1 && treeList.get(root).getRight() != -1){
			bulidhuftreecode(treeList.get(root).getLeft(), str+"0");
			bulidhuftreecode(treeList.get(root).getRight(), str + "1");
		}
		else{
			treeList.get(root).code = str;
		}	
	}

java实现哈夫曼压缩与解压缩的方法

4.哈夫曼编码代替字符,进行压缩,压缩前首先要将文件头(文件标志,字符数量,最后一个字节有效位,密码)字符和其频率的那张表格写入文件,以便于解压缩

public void creatCodeFile(String path) throws Exception{
		byte value = 0;
		int index = 0;
		int arr[] = new int[256];
		int intchar;
		
		for(int i = 0; i < charCount; i++){
			arr[treeList.get(i).freq.character] = i;
			
		}
		File file = new File(path);
    if(!file.exists()){
       if(!file.createNewFile()){
      	 throw new Exception("创建文件失败");
       }
    }
		int count = charList.size();
		HuffmanHead head = new HuffmanHead(count, howlongchar(count), password);
        //将文件头信息写入文件
		this.write = new RandomAccessFile(file, "rw");
		write.write(head.InfoToByte());
        //将字符及其频率的表写入文件
		for(HuffmanFreq freq : charList){
			byte[] bt = freq.infoToByte();
			write.write(bt);
		}
		//将字符用哈夫曼编码进行压缩,这里读写都是采用缓存机制
		byte[] readBuffer = new byte[BUFFER_SIZE];
		while((intchar = read.read(readBuffer))!= -1){
			ProgressBar.SetCurrent(read.getFilePointer());
			for(int i = 0; i < intchar;i++){
				int temp = readBuffer[i]& 0xff; 
				String code = treeList.get(arr[temp]).code;
				char[] chars = code.toCharArray();
				
				for(int j = 0; j < chars.length; j++){
					if(chars[j] == '0'){
						value = CLR_BYTE(value, index);
					}
					if(chars[j] == '1'){
						value = SET_BYTE(value, index);
					}
					if(++index >= 8){
						index = 0;
						writeInBuffer(value);
					}
				}
			}
		}
		//此方法速度较慢
//		while((intchar = is.read()) != -1){
//			String code = treeList.get(arr[intchar]).code;
//			char[] chars = code.toCharArray();
//			
//			for(int i = 0; i < chars.length; i++){
//				if(chars[i] == '0'){
//					value = CLR_BYTE(value, index);
//				}
//				if(chars[i] == '1'){
//					value = SET_BYTE(value, index);
//				}
//				if(++index >= 8){
//					index = 0;
//					oos.write(value);
//				}
//			}
//		}
		if(index != 0){
			writeInBuffer(value);
		}
	  byte[] Data = Arrays.copyOfRange(writeBuffer, 0, writeBufferSize);
	  write.write(Data);
	  write.close();
		read.close();
	}
    //指定位,置1
    byte SET_BYTE(byte value, int index){
		return (value) |= (1 << ((index) ^ 7));
	}	
    //指定位,置0
	byte CLR_BYTE(byte value, int index){ 
		return (value) &= (~(1 << ((index) ^ 7)));
	}
    //判断指定位是否为0,0为false,1为true
	boolean GET_BYTE(byte value, int index){ 
		return ((value) & (1 << ((index) ^ 7))) != 0;
	}

如果一个字节一个字节往文件里写,速度会极慢,为了提高效率,写也采用缓存,先写到缓存区,缓存区满了后写入文件,

    private void writeInBuffer(byte value) throws Exception {
		if(writeBufferSize < BUFFER_SIZE){
			writeBuffer[writeBufferSize] = value;
			if(++writeBufferSize >= BUFFER_SIZE){
				write.write(writeBuffer);
				writeBufferSize = 0;
			}
		} else{
			throw new Exception("写入文件出错");
		}
	}

到这里压缩就完成了,以下为解压缩方法

1.从写入文件中的字符统计的表读出放入list里

public void init() throws Exception{
		char isHUf = read.readChar();
        //验证文件头信息
		if(isHUf != '哈'){
			throw new Exception("该文件不是HUFFMAN压缩文件");
		}
		this.charCount = read.readChar();
		this.treeSize = 2*charCount -1;
		this.lastIndex = read.readChar();
		int password = read.readInt();
		if(password != this.password.hashCode()){
			System.out.println("密码错误");
		} else{
			System.out.println("密码正确,正在解压");
		}
		
        //从文件中将字符统计的表读出
		byte[] buffer = new byte[charCount * 6];
		read.seek(10);
		read.read(buffer, 0, charCount * 6);
		ProgressBar.SetCurrent(read.getFilePointer());
		for(int i = 0; i < buffer.length; i+=6){
			byte[] buff = Arrays.copyOfRange(buffer, i, i+2);
			ByteBuffer bb = ByteBuffer.allocate (buff.length);
		  bb.put (buff);
		  bb.flip ();
		  CharBuffer cb = cs.decode (bb);
		  byte[] buff1 = Arrays.copyOfRange(buffer, i+2, i+6);
		  int size = ByteAnd8Types.bytesToInt2(buff1, 0);
		  HuffmanFreq freq = new HuffmanFreq(cb.array()[0], size);
		  charList.add(freq);
		}
	}

2.用统计结果构建哈夫曼树(和以上代码一样)

3.用哈夫曼树生成哈夫曼编码(从根结点开始,路径左边记为0,右边记为1)(和以上代码一样)

4.遍历文件每个字节,根据哈夫曼编码找到对应的字符,将字符写入新文件

 public void creatsourcefile(String pathname) throws Exception{
		int root = treeList.size() - 1;
		int fininsh = 1;
		long len;
		File file = new File(pathname);
		if(!file.exists()){
			 if(!file.createNewFile()){
				 throw new Exception("创建文件失败");
	     }
		}
		write = new RandomAccessFile(file, "rw");
		
		int intchar;
		byte[] bytes = new byte[1<<18];
		int index = 0;
		while((intchar = read.read(bytes))!= -1){
			len = read.getFilePointer();
			ProgressBar.SetCurrent(len);
			for(int i = 0; i < intchar;i++){
				for(;index < 8 && fininsh != 0;){
					if(GET_BYTE(bytes[i], index)){
						root = treeList.get(root).right;
					} else{
						root = treeList.get(root).left;
					}
					if(treeList.get(root).right== -1 && treeList.get(root).left == -1){
						byte temp = (byte)treeList.get(root).freq.character;
						writeInBuffer(temp);
						root = treeList.size() - 1;
					}
					index++;
					if(len == this.goalfilelenth && i == intchar-1){
						if(index >= this.lastIndex){
							fininsh = 0;
						}
					}
				}
				index = 0;
			}
		}
		byte[] Data = Arrays.copyOfRange(writeBuffer, 0, writeBufferSize);
		write.write(Data);
		write.close();
		write.close();
		read.close();
	}

四运行展示:

java实现哈夫曼压缩与解压缩的方法

java实现哈夫曼压缩与解压缩的方法

以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持创新互联。


本文题目:java实现哈夫曼压缩与解压缩的方法
本文地址:http://myzitong.com/article/iedoeg.html