applybox.cpp

来自「一个google的OCR源码」· C++ 代码 · 共 892 行 · 第 1/3 页

/**********************************************************************
 * File:        applybox.cpp  (Formerly applybox.c)
 * Description: Re segment rows according to box file data
 * Author:		Phil Cheatle
 * Created:		Wed Nov 24 09:11:23 GMT 1993
 *
 * (C) Copyright 1993, Hewlett-Packard Ltd.
 ** Licensed under the Apache License, Version 2.0 (the "License");
 ** you may not use this file except in compliance with the License.
 ** You may obtain a copy of the License at
 ** http://www.apache.org/licenses/LICENSE-2.0
 ** Unless required by applicable law or agreed to in writing, software
 ** distributed under the License is distributed on an "AS IS" BASIS,
 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 ** See the License for the specific language governing permissions and
 ** limitations under the License.
 *
 **********************************************************************/
/*
define SECURE_NAMES for code versions which go to UNLV to stop tessedit
including all the newdiff stuff (which contains lots of text indicating
what measures we are interested in.
*/
/* #define SECURE_NAMES done in secnames.h when necessary*/

#include "mfcpch.h"
#include "applybox.h"
#include <ctype.h>
#include <string.h>
#ifdef __UNIX__
#include <assert.h>
#include <errno.h>
#endif
#include "boxread.h"
#include "mainblk.h"
#include "genblob.h"
#include "fixxht.h"
#include "control.h"
#include "tessbox.h"
#include "globals.h"
#include "secname.h"
#include "unichar.h"
#include "matchdefs.h"

#define SECURE_NAMES
#ifndef SECURE_NAMES
#include          "wordstats.h"
#endif

#define EXTERN
EXTERN BOOL_VAR (applybox_rebalance, TRUE, "Drop dead");
EXTERN INT_VAR (applybox_debug, 5, "Debug level");
EXTERN INT_VAR (applybox_page, 0, "Page number to apply boxes from");
EXTERN STRING_VAR (applybox_test_exclusions, "",
"Chars ignored for testing");
EXTERN double_VAR (applybox_error_band, 0.15, "Err band as fract of xht");

// The unicharset used during box training
static UNICHARSET unicharset_boxes;

static void PrintString(const char* str) {
  tprintf("%s:", str);
  int step = 0;
  for (int i = 0; str[i]; i += step) {
    step = UNICHAR::utf8_step(str + i);
    if (step == 0)
      step = 1;
    UNICHAR ch(str + i, step);
    tprintf("[%x]", ch.first_uni());
  }
  tprintf("\n", str);
}

/*************************************************************************
 * The code re-assigns outlines to form words each with ONE labelled blob.
 * Noise is left in UNLABELLED words. The chars on the page are checked crudely
 * for sensible position relative to baseline and xht. Failed boxes are
 * compensated for by duplicating other believable instances of the character.
 *
 * The box file is assumed to contain box definitions, one per line, of the
 * following format:
 *   <Char> <left> <bottom> <right> <top> ... arbitrary trailing fields unused
 *
 * The approach taken is to search the WHOLE page for stuff overlapping each box.
 *	- This is not too inefficient and is SAFE.
 *    - We can detect overlapping blobs as we will be attempting to put a blob
 *      from a LABELLED word into the current word.
 *    - When all the boxes have been processed we can detect any stuff which is
 *      being ignored - it is the unlabelled words left on the page.
 *
 * A box should only overlap one row.
 *
 * A warning is given if the box is on the same row as the previous box, but NOT
 * on the same row as the previous blob.
 *
 * Any OUTLINE which overlaps the box is put into the new word.
 *
 * ascender chars must ascend above xht significantly
 * xht chars must not rise above row xht significantly
 * bl chars must not descend below baseline significantly
 * descender chars must descend below baseline significantly
 *
 * ?? Certain chars are DROPPED - to limit the training data.
 *
 *************************************************************************/

void apply_boxes(BLOCK_LIST *block_list    //real blocks
                ) {
  inT16 boxfile_lineno = 0;
  inT16 boxfile_charno = 0;
  TBOX box;                       //boxfile box
  UNICHAR_ID uch_id;             //correct ch from boxfile
  ROW *row;
  ROW *prev_row = NULL;
  inT16 prev_box_right = MAX_INT16;
  inT16 block_id;
  inT16 row_id;
  inT16 box_count = 0;
  inT16 box_failures = 0;
  inT16 labels_ok;
  inT16 rows_ok;
  inT16 bad_blobs;
  inT16 tgt_char_counts[MAX_NUM_CLASSES];    //No. of box samples
  //      inT16                                   labelled_char_counts[128];      //No. of unique labelled samples
  inT16 i;
  inT16 rebalance_count = 0;
  UNICHAR_ID min_uch_id;
  inT16 min_samples;
  inT16 final_labelled_blob_count;

  // Clean the unichar set
  unicharset_boxes.clear();
  // Space character needed to represent NIL classification
  unicharset_boxes.unichar_insert(" ");

  for (i = 0; i < MAX_NUM_CLASSES; i++)
    tgt_char_counts[i] = 0;

  FILE* box_file;
  STRING filename = imagefile;
  filename += ".box";
  if (!(box_file = fopen (filename.string(), "r"))) {
    CANTOPENFILE.error ("read_next_box", EXIT,
      "Cant open box file %s %d",
      filename.string(), errno);
  }

  clear_any_old_text(block_list);
  while (read_next_box(applybox_page, box_file, &box, &uch_id)) {
    box_count++;
    tgt_char_counts[uch_id]++;
    row = find_row_of_box (block_list, box, block_id, row_id);
    if (box.left () < prev_box_right) {
      boxfile_lineno++;
      boxfile_charno = 1;
    }
    else
      boxfile_charno++;

    if (row == NULL) {
      box_failures++;
      report_failed_box (boxfile_lineno, boxfile_charno, box,
                         unicharset_boxes.id_to_unichar(uch_id),
        "FAILURE! box overlaps no blobs or blobs in multiple rows");
    }
    else {
      if ((box.left () >= prev_box_right) && (row != prev_row))
        report_failed_box (boxfile_lineno, boxfile_charno, box,
                           unicharset_boxes.id_to_unichar(uch_id),
          "WARNING! false row break");
      box_failures += resegment_box (row, box, uch_id, block_id, row_id,
        boxfile_lineno, boxfile_charno);
      prev_row = row;
    }
    prev_box_right = box.right ();
  }
  tidy_up(block_list,
          labels_ok,
          rows_ok,
          bad_blobs,
          tgt_char_counts,
          rebalance_count,
          &min_uch_id,
          min_samples,
          final_labelled_blob_count);
  tprintf ("APPLY_BOXES:\n");
  tprintf ("   Boxes read from boxfile:  %6d\n", box_count);
  tprintf ("   Initially labelled blobs: %6d in %d rows\n",
    labels_ok, rows_ok);
  tprintf ("   Box failures detected:		%6d\n", box_failures);
  tprintf ("   Duped blobs for rebalance:%6d\n", rebalance_count);
  tprintf ("   \"%s\" has fewest samples:%6d\n",
           unicharset_boxes.id_to_unichar(min_uch_id), min_samples);
  tprintf ("				Total unlabelled words:   %6d\n",
    bad_blobs);
  tprintf ("				Final labelled words:     %6d\n",
    final_labelled_blob_count);
}


void clear_any_old_text(                        //remove correct text
                        BLOCK_LIST *block_list  //real blocks
                       ) {
  BLOCK_IT block_it(block_list);
  ROW_IT row_it;
  WERD_IT word_it;

  for (block_it.mark_cycle_pt ();
  !block_it.cycled_list (); block_it.forward ()) {
    row_it.set_to_list (block_it.data ()->row_list ());
    for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
      word_it.set_to_list (row_it.data ()->word_list ());
      for (word_it.mark_cycle_pt ();
      !word_it.cycled_list (); word_it.forward ()) {
        word_it.data ()->set_text ("");
      }
    }
  }
}


BOOL8 read_next_box(int page,
                    FILE* box_file,  //
                    TBOX *box,
                    UNICHAR_ID *uch_id) {
  int x_min;
  int y_min;
  int x_max;
  int y_max;
  char uch[kBoxReadBufSize];

  while (read_next_box(page, box_file, uch, &x_min, &y_min, &x_max, &y_max)) {
    if (!unicharset_boxes.contains_unichar(uch))
    {
      unicharset_boxes.unichar_insert(uch);
      if (unicharset_boxes.size() > MAX_NUM_CLASSES) {
        tprintf("Error: Size of unicharset of boxes is "
                "greater than MAX_NUM_CLASSES (%d)\n",
                MAX_NUM_CLASSES);
        exit(1);
      }
    }
    *uch_id = unicharset_boxes.unichar_to_id(uch);
    *box = TBOX (ICOORD (x_min, y_min), ICOORD (x_max, y_max));
    return TRUE;             //read a box ok
  }
  return FALSE;                  //EOF
}


ROW *find_row_of_box(                         //
                     BLOCK_LIST *block_list,  //real blocks
                     TBOX box,                 //from boxfile
                     inT16 &block_id,
                     inT16 &row_id_to_process) {
  BLOCK_IT block_it(block_list);
  BLOCK *block;
  ROW_IT row_it;
  ROW *row;
  ROW *row_to_process = NULL;
  inT16 row_id;
  WERD_IT word_it;
  WERD *word;
  BOOL8 polyg;
  PBLOB_IT blob_it;
  PBLOB *blob;
  OUTLINE_IT outline_it;
  OUTLINE *outline;

  /*
    Find row to process - error if box REALLY overlaps more than one row. (I.e
    it overlaps blobs in the row - not just overlaps the bounding box of the
    whole row.)
  */

  block_id = 0;
  for (block_it.mark_cycle_pt ();
  !block_it.cycled_list (); block_it.forward ()) {
    block_id++;
    row_id = 0;
    block = block_it.data ();
    if (block->bounding_box ().overlap (box)) {
      row_it.set_to_list (block->row_list ());
      for (row_it.mark_cycle_pt ();
      !row_it.cycled_list (); row_it.forward ()) {
        row_id++;
        row = row_it.data ();
        if (row->bounding_box ().overlap (box)) {
          word_it.set_to_list (row->word_list ());
          for (word_it.mark_cycle_pt ();
          !word_it.cycled_list (); word_it.forward ()) {
            word = word_it.data ();
            polyg = word->flag (W_POLYGON);
            if (word->bounding_box ().overlap (box)) {
              blob_it.set_to_list (word->gblob_list ());
              for (blob_it.mark_cycle_pt ();
              !blob_it.cycled_list (); blob_it.forward ()) {
                blob = blob_it.data ();