java - JPEG如何在解码文件流时跳过用户定义的标签

Question

public ArrayList DCTread(char[] im,int flag,int select, int DC0,int row,int col){
    //Input:im is the binary sequence of the host image. I wrote a "byte2char" function to convert that. flag serves as an outside pointer to locate the to-be-decoded chars. DC0 is the DC coeff for the last block.And row,col is simply for debug.
    //Main Output:An ArrayList that contains the DCT coeffs,pointer(an int showing how many bits are read in this function)
    String rev=new String();char[] DCcode;
    ArrayList res = new ArrayList(2);int[][] ac = null; int[][] dc = null;int[][] coeff = new int[8][8];int pointer = 0;
    int[] ans;int wordLen;int zeroLen;int diff;int ACnum = 1;int dct;
    switch(select){//determine using which two huffman trees.
    case(0):ac = a0;dc = d0;break;
    case(1):ac = a1;dc = d0;break;
    case(16):ac = a0;dc = d1;break;
    case(17):ac = a1;dc = d1;break;
    }
    //DC
    ans = T.huffmanDecoder(im,pointer+flag,dc,row,col);
    if(ans[0]==-1){
        int a1 = T.bin2dec_str(im,pointer+flag,8);int a2 = T.bin2dec_str(im,pointer+flag+8,8);
        pointer +=16;//I wish to skip the User Defined Tags by reading its length
        int autoLen = T.bin2dec_str(im,pointer+flag,8)*16+T.bin2dec_str(im,pointer+flag+8,8);
        pointer +=autoLen*8;
    }
    ans = T.huffmanDecoder(im,pointer+flag,dc,row,col);
    pointer += ans[0];wordLen = ans[1];
    diff = T.i_unsignDecoder(T.bin2dec_str(im,pointer+flag,wordLen),wordLen);   
    coeff[0][0]= DC0 + diff;
    pointer += wordLen;DCcode=Arrays.copyOfRange(im, flag, pointer+flag);
    //AC
    while(ACnum<=63){
        ans = T.huffmanDecoder(im,pointer+flag,ac,row,col);
        pointer += ans[0];
        if(ans[1]==0){//
            break;}
        zeroLen = (ans[1]&(0xF0))/16;wordLen = ans[1]&(0x0F);
        for(int j=0;j<zeroLen;j++){
            coeff[zigZag[ACnum][0]][zigZag[ACnum][1]] = 0;
                    ACnum ++;
        }
        dct = T.i_unsignDecoder(T.bin2dec_str(im,pointer+flag,wordLen),wordLen);
        pointer += wordLen;

        coeff[zigZag[ACnum][0]][zigZag[ACnum][1]] = dct;
        ACnum ++;
    }

    res.add(coeff);
    res.add(pointer);
    res.add(DCcode);
    return res;

}

大家好，首先我很高兴欢迎您看到困扰我两天的棘手问题，并非常感谢您在百忙之中帮助我解决这个问题。我已经关注 StackOverflow 很长时间了，但这确实是我第一次提出问题。

我想要的是在 Java 中读取 JPEG 的 DCT，而不使用 libjpeg 库(用 C++ 编写)。但是，我遇到了许多用户定义的标签(UDT)，我发现使用上面在算法中列出的方法很难跳过它们。我对 UDT 不太熟悉。

它们不是以“0xFFXX 0x....”开头(其中 0x.... 给出了该标签的长度)？您的建议将对我有很大帮助。谢谢!

Answer 1

可以由用户定义的标记是 APPn 和 COM。这些标记后面是 BIG ENDIAN 格式的长度。

但是，我很惊讶您发现“很多”这样的标签。通常，JPEG 流中只有一两个。