c++ - 如何将位图转换为内存中的 PIX？

Question

Tesseract 似乎不能很好地处理位图，它可以处理某些输入但会搞砸其他输入。同时在与之前相同但采用 leptonica 格式 PIX 的输入上表现良好。

如何将内存中的位图转换为 PIX？

想到的一个想法是使用 leptonica 的 pixReadMem() :

00724 /*---------------------------------------------------------------------*
00725  *                            Read from memory                         *
00726  *---------------------------------------------------------------------*/
00727 /*!
00728  *  pixReadMem()
00729  *
00730  *      Input:  data (const; encoded)
00731  *              datasize (size of data)
00732  *      Return: pix, or null on error
00733  *
00734  *  Notes:
00735  *      (1) This is a variation of pixReadStream(), where the data is read
00736  *          from a memory buffer rather than a file.
00737  *      (2) On windows, this will only read tiff formatted files from
00738  *          memory.  For other formats, it requires fmemopen(3).
00739  *          Attempts to read those formats will fail at runtime.
00740  *      (3) findFileFormatBuffer() requires up to 8 bytes to decide on
00741  *          the format.  That determines the constraint here.
00742  */

所以现在我需要找到一种方法来从内存中的位图在内存中构造一个 TIFF。但我不知道怎么办。

Answer 1

好吧，我决定自己用一个快速的脏端口来做这件事。我只是使用文件加载 BMP 函数并创建了假的 fread/fseek，因为我真的很懒。有效。

如果有人需要，我会贴在下面。令人困惑的是，为什么 leptonica 开发人员没有为 Windows 正确地做到这一点。

#include <cstring>

#include <leptonica/allheaders.h>
#include "leptonica_hack.h"
#include "bmp.h"

int fake_file_tracker = 0;

size_t fake_fread ( void * ptr, size_t size, size_t count, char * buffer, size_t buflen)
{
    if (fake_file_tracker >= buflen)
    {
        return 0;
    }

    if (fake_file_tracker + size * count > buflen)
    {
        (void)memcpy(ptr, (void *)(buffer+fake_file_tracker), (size_t)(buflen - fake_file_tracker));
        fake_file_tracker = buflen+1;
        return (size_t)((buflen - fake_file_tracker) / size);
    }

    (void)memcpy(ptr, (void *)(buffer+fake_file_tracker), size * count);
    fake_file_tracker += size * count;
    return count;

}

int fake_fseek ( char * buffer, long int offset, int origin, size_t buflen)
{
    if (origin + offset >= buflen)
        return 0;

    fake_file_tracker = origin + offset;
    return 0;
}


/*!
 *  pixReadStreamBmp()
 *
 *      Input:  stream opened for read
 *      Return: pix, or null on error
 *
 *  Notes:
 *      (1) Here are references on the bmp file format:
 *          http://en.wikipedia.org/wiki/BMP_file_format
 *          http://www.fortunecity.com/skyscraper/windows/364/bmpffrmt.html
 */
PIX *
pixReadBmpFromBuffer(char  *fp, size_t buflen)
{
    fake_file_tracker = 0;
l_uint16   sval;
l_uint32   ival;
l_int16    bfType, bfSize, bfFill1, bfReserved1, bfReserved2;
l_int16    offset, bfFill2, biPlanes, depth, d;
l_int32    biSize, width, height, xres, yres, compression, ignore;
l_int32    imagebytes, biClrUsed, biClrImportant;
l_uint8   *colormapBuf = NULL;
l_int32    colormapEntries;
l_int32    fileBpl, extrabytes, readerror;
l_int32    pixWpl, pixBpl;
l_int32    i, j, k;
l_uint8    pel[4];
l_uint8   *data;
l_uint32  *line, *pword;
PIX        *pix, *pixt;
PIXCMAP   *cmap;

    PROCNAME("pixReadBmpFromBuffer");

    if (!fp)
        return (PIX *)ERROR_PTR("fp not defined", procName, NULL);

        /* Read bitmap file header */
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    bfType = convertOnBigEnd16(sval);
    if (bfType != BMP_ID)
        return (PIX *)ERROR_PTR("not bmf format", procName, NULL);

    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    bfSize = convertOnBigEnd16(sval);
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    bfFill1 = convertOnBigEnd16(sval);
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    bfReserved1 = convertOnBigEnd16(sval);
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    bfReserved2 = convertOnBigEnd16(sval);
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    offset = convertOnBigEnd16(sval);
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    bfFill2 = convertOnBigEnd16(sval);

        /* Read bitmap info header */
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    biSize = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    width = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    height = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    biPlanes = convertOnBigEnd16(sval);
    ignore = fake_fread((char *)&sval, 1, 2, fp, buflen);
    depth = convertOnBigEnd16(sval);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    compression = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    imagebytes = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    xres = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    yres = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    biClrUsed = convertOnBigEnd32(ival);
    ignore = fake_fread((char *)&ival, 1, 4, fp, buflen);
    biClrImportant = convertOnBigEnd32(ival);

    if (compression != 0)
        return (PIX *)ERROR_PTR("cannot read compressed BMP files",
                                procName,NULL);

        /* A little sanity checking.  It would be nice to check
         * if the number of bytes in the file equals the offset to
         * the data plus the imagedata, but this won't work when
         * reading from memory, because fmemopen() doesn't implement
         * ftell().  So we can't do that check.  The imagebytes for
         * uncompressed images is either 0 or the size of the file data.
         * (The fact that it can be 0 is perhaps some legacy glitch).  */
    if (width < 1)
        return (PIX *)ERROR_PTR("width < 1", procName,NULL);
    if (height < 1)
        return (PIX *)ERROR_PTR("height < 1", procName,NULL);
    if (depth < 1 || depth > 32)
        return (PIX *)ERROR_PTR("depth not in [1 ... 32]", procName,NULL);
    fileBpl = 4 * ((width * depth + 31)/32);
    if (imagebytes != 0 && imagebytes != fileBpl * height)
        return (PIX *)ERROR_PTR("invalid imagebytes", procName,NULL);
    if (offset < BMP_FHBYTES + BMP_IHBYTES)
        return (PIX *)ERROR_PTR("invalid offset: too small", procName,NULL);
    if (offset > BMP_FHBYTES + BMP_IHBYTES + 4 * 256)
        return (PIX *)ERROR_PTR("invalid offset: too large", procName,NULL);

        /* Handle the colormap */
    colormapEntries = (offset - BMP_FHBYTES - BMP_IHBYTES) / sizeof(RGBA_QUAD);
    if (colormapEntries > 0) {
        if ((colormapBuf = (l_uint8 *)CALLOC(colormapEntries,
                                             sizeof(RGBA_QUAD))) == NULL)
            return (PIX *)ERROR_PTR("colormapBuf alloc fail", procName, NULL );

            /* Read colormap */
        if (fake_fread(colormapBuf, sizeof(RGBA_QUAD), colormapEntries, fp, buflen)
                 != colormapEntries) {
            FREE(colormapBuf);
            return (PIX *)ERROR_PTR( "colormap read fail", procName, NULL);
        }
    }

        /* Make a 32 bpp pix if depth is 24 bpp */
    d = depth;
    if (depth == 24)
        d = 32;
    if ((pix = pixCreate(width, height, d)) == NULL)
        return (PIX *)ERROR_PTR( "pix not made", procName, NULL);
    pixSetXRes(pix, (l_int32)((l_float32)xres / 39.37 + 0.5));  /* to ppi */
    pixSetYRes(pix, (l_int32)((l_float32)yres / 39.37 + 0.5));  /* to ppi */
    pixWpl = pixGetWpl(pix);
    pixBpl = 4 * pixWpl;

    cmap = NULL;
    if (colormapEntries > 256)
        L_WARNING("more than 256 colormap entries!", procName);
    if (colormapEntries > 0) {  /* import the colormap to the pix cmap */
        cmap = pixcmapCreate(L_MIN(d, 8));
        FREE(cmap->array);  /* remove generated cmap array */
        cmap->array  = (void *)colormapBuf;  /* and replace */
        cmap->n = L_MIN(colormapEntries, 256);
    }
    pixSetColormap(pix, cmap);

        /* Seek to the start of the bitmap in the file */
    fake_fseek(fp, offset, 0, buflen);

    if (depth != 24) {  /* typ. 1 or 8 bpp */
        data = (l_uint8 *)pixGetData(pix) + pixBpl * (height - 1);
        for (i = 0; i < height; i++) {
            if (fake_fread(data, 1, fileBpl, fp, buflen) != fileBpl) {
                pixDestroy(&pix);
                return (PIX *)ERROR_PTR("BMP read fail", procName, NULL);
            }
            data -= pixBpl;
        }
    }
    else {  /*  24 bpp file; 32 bpp pix
             *  Note: for bmp files, pel[0] is blue, pel[1] is green,
             *  and pel[2] is red.  This is opposite to the storage
             *  in the pix, which puts the red pixel in the 0 byte,
             *  the green in the 1 byte and the blue in the 2 byte.
             *  Note also that all words are endian flipped after
             *  assignment on L_LITTLE_ENDIAN platforms.
             *
             *  We can then make these assignments for little endians:
             *      SET_DATA_BYTE(pword, 1, pel[0]);      blue
             *      SET_DATA_BYTE(pword, 2, pel[1]);      green
             *      SET_DATA_BYTE(pword, 3, pel[2]);      red
             *  This looks like:
             *          3  (R)     2  (G)        1  (B)        0
             *      |-----------|------------|-----------|-----------|
             *  and after byte flipping:
             *           3          2  (B)     1  (G)        0  (R)
             *      |-----------|------------|-----------|-----------|
             *
             *  For big endians we set:
             *      SET_DATA_BYTE(pword, 2, pel[0]);      blue
             *      SET_DATA_BYTE(pword, 1, pel[1]);      green
             *      SET_DATA_BYTE(pword, 0, pel[2]);      red
             *  This looks like:
             *          0  (R)     1  (G)        2  (B)        3
             *      |-----------|------------|-----------|-----------|
             *  so in both cases we get the correct assignment in the PIX.
             *
             *  Can we do a platform-independent assignment?
             *  Yes, set the bytes without using macros:
             *      *((l_uint8 *)pword) = pel[2];           red
             *      *((l_uint8 *)pword + 1) = pel[1];       green
             *      *((l_uint8 *)pword + 2) = pel[0];       blue
             *  For little endians, before flipping, this looks again like:
             *          3  (R)     2  (G)        1  (B)        0
             *      |-----------|------------|-----------|-----------|
             */
        readerror = 0;
        extrabytes = fileBpl - 3 * width;
        line = pixGetData(pix) + pixWpl * (height - 1);
        for (i = 0; i < height; i++) {
            for (j = 0; j < width; j++) {
                pword = line + j;
                if (fake_fread(&pel, 1, 3, fp, buflen) != 3)
                    readerror = 1;
                *((l_uint8 *)pword + COLOR_RED) = pel[2];
                *((l_uint8 *)pword + COLOR_GREEN) = pel[1];
                *((l_uint8 *)pword + COLOR_BLUE) = pel[0];
            }
            if (extrabytes) {
                for (k = 0; k < extrabytes; k++)
                    ignore = fake_fread(&pel, 1, 1, fp, buflen);
            }
            line -= pixWpl;
        }
        if (readerror) {
            pixDestroy(&pix);
            return (PIX *)ERROR_PTR("BMP read fail", procName, NULL);
        }
    }

    pixEndianByteSwap(pix);

        /* ----------------------------------------------
         * The bmp colormap determines the values of black
         * and white pixels for binary in the following way:
         * if black = 1 (255), white = 0
         *      255, 255, 255, 0, 0, 0, 0, 0
         * if black = 0, white = 1 (255)
         *      0, 0, 0, 0, 255, 255, 255, 0
         * We have no need for a 1 bpp pix with a colormap!
         * ---------------------------------------------- */
    if (depth == 1 && cmap) {
/*        L_INFO("Removing colormap", procName); */
        pixt = pixRemoveColormap(pix, REMOVE_CMAP_BASED_ON_SRC);
        pixDestroy(&pix);
        pix = pixt;  /* rename */
    }

    return pix;
}