control.cpp

来自「一个google的OCR源码」· C++ 代码 · 共 1,843 行 · 第 1/5 页

                         WERD_RES *word,
                         ROW *row) {
  BOOL8 done_this_pass = FALSE;
  WERD_RES new_x_ht_word (word->word);
  float new_x_ht = 0.0;
  inT16 old_xht_reject_count;
  inT16 new_xht_reject_count;
  inT16 old_xht_accept_count;
  inT16 new_xht_accept_count;
  BOOL8 accept_new_x_ht = FALSE;
  inT16 old_chs_in_wd;
  inT16 new_chs_in_wd;
  inT16 old_word_quality;
  inT16 new_word_quality;
  inT16 dummy;

  set_global_subloc_code(SUBLOC_NORM);
  check_debug_pt (word, 30);
  if (!word->done ||
    tessedit_training_tess ||
  tessedit_training_wiseowl || tessedit_dump_choices) {
    word->caps_height = 0.0;
    if (word->x_height == 0.0f)
      word->x_height = row->x_height();
    if (word->outword != NULL) {
      delete word->outword;      //get rid of junk
      delete word->best_choice;
      delete word->raw_choice;
    }
    match_word_pass2 (word, row, word->x_height);
    done_this_pass = TRUE;
    check_debug_pt (word, 40);
  }

  if (!word->tess_failed && !word->word->flag (W_REP_CHAR)) {
    set_global_subloc_code(SUBLOC_FIX_XHT);
    if ((tessedit_xht_fiddles_on_done_wds || !word->done) &&
      (tessedit_xht_fiddles_on_no_rej_wds ||
    (word->reject_map.reject_count () > 0))) {
      if ((x_ht_check_word_occ >= 2) && word_occ_first)
        check_block_occ(word);

      if (tessedit_redo_xheight)
        re_estimate_x_ht(word, &new_x_ht);

      if (((x_ht_check_word_occ >= 2) && !word_occ_first) ||
        ((x_ht_check_word_occ >= 1) && (new_x_ht > 0)))
        check_block_occ(word);
    }
    if (new_x_ht > 0) {
      old_chs_in_wd = word->reject_map.length ();

      /* Re-estimated x_ht error suggests a rematch is worthwhile. */
      new_x_ht_word.x_height = new_x_ht;
      new_x_ht_word.caps_height = 0.0;
      match_word_pass2 (&new_x_ht_word, row, new_x_ht_word.x_height);
      if (!new_x_ht_word.tess_failed) {
        if ((x_ht_check_word_occ >= 1) && word_occ_first)
          check_block_occ(&new_x_ht_word);

        re_estimate_x_ht(&new_x_ht_word, &new_x_ht);

        if ((x_ht_check_word_occ >= 1) && !word_occ_first)
          check_block_occ(&new_x_ht_word);

        old_xht_reject_count = word->reject_map.reject_count ();
        old_xht_accept_count = old_chs_in_wd - old_xht_reject_count;
        new_xht_reject_count = new_x_ht_word.reject_map.reject_count ();
        new_chs_in_wd = new_x_ht_word.reject_map.length ();
        new_xht_accept_count = new_chs_in_wd - new_xht_reject_count;
        accept_new_x_ht =
          ((new_xht_accept_count > old_xht_accept_count) ||
          ((new_xht_accept_count == old_xht_accept_count) &&
          (new_xht_accept_count > 0))) &&
          (!new_x_ht_word.guessed_x_ht ||
          !new_x_ht_word.guessed_caps_ht);

        if (accept_new_x_ht && x_ht_quality_check) {
          word_char_quality(word, row, &old_word_quality, &dummy);
          word_char_quality(&new_x_ht_word, row, &new_word_quality, &dummy);
          if (old_word_quality > new_word_quality)
            accept_new_x_ht = FALSE;
        }

        if (accept_new_x_ht && (x_ht_stringency > 0)) {
          accept_new_x_ht =
            (count_alphanums (&new_x_ht_word) > x_ht_stringency);
          if (!accept_new_x_ht && rej_use_xht) {
            if (debug_x_ht_level >= 1)
              tprintf
                ("Failed stringency test so reject original word\n");
            word->reject_map.rej_word_xht_fixup ();
          }
        }

        #ifndef SECURE_NAMES
        if (debug_x_ht_level >= 1) {
          tprintf ("New XHT Match:: %s ",
            word->best_choice->string ().string ());
          word->reject_map.print (debug_fp);
          tprintf (" -> %s ",
            new_x_ht_word.best_choice->string ().string ());
          new_x_ht_word.reject_map.print (debug_fp);
          tprintf (" %s->%s %s %s\n",
            word->guessed_x_ht ? "GUESS" : "CERT",
            new_x_ht_word.guessed_x_ht ? "GUESS" : "CERT",
            new_x_ht > 0.1 ? "STILL DOUBT" : "OK",
            accept_new_x_ht ? "ACCEPTED" : "");
        }
        #endif
      }
      if (accept_new_x_ht) {
        /*
           The new x_ht is deemed superior so put the final results in the real word
           and destroy the old results
         */
        delete word->outword;    //get rid of junk
        word->outword = new_x_ht_word.outword;
        word->denorm = new_x_ht_word.denorm;
        delete word->best_choice;
        word->best_choice = new_x_ht_word.best_choice;
        delete word->raw_choice;
        word->raw_choice = new_x_ht_word.raw_choice;
        word->reject_map = new_x_ht_word.reject_map;
        word->done = new_x_ht_word.done;
        done_this_pass = TRUE;
      }
      else {
      /*
         The new x_ht is no better, so destroy the copy word and put any uncertain
         x or cap ht estimate back to default. (I.e. dont blame me if its bad!)
         Conditionally, use any ammended block occ chars.
       */
                                 //get rid of junk
        delete new_x_ht_word.outword;
        delete new_x_ht_word.best_choice;
        delete new_x_ht_word.raw_choice;
      }
                                 //to keep new destructor happy
      new_x_ht_word.outword = NULL;
                                 //to keep new destructor happy
      new_x_ht_word.best_choice = NULL;
                                 //to keep new destructor happy
      new_x_ht_word.raw_choice = NULL;

      if (rej_mostly_reject_mode == 2) {
        reject_mostly_rejects(word);
        tprintf ("Rejecting mostly rejects on %s ",
          word->best_choice->string ().string ());
      }
    }

    set_global_subloc_code(SUBLOC_NORM);

    if (done_this_pass && !word->done && tessedit_save_stats)
      SaveBadWord (word->best_choice->string ().string (),
        word->best_choice->certainty ());
    record_certainty (word->best_choice->certainty (), 2);
    //accounting
  }
#ifndef GRAPHICS_DISABLED
  if (tessedit_draw_outwords) {
    if (fx_win == NULL)
      create_fx_win();
    clear_fx_win();
    word->outword->plot (fx_win);
    TBOX wbox = word->outword->bounding_box();
    fx_win->ZoomToRectangle(wbox.left(), wbox.top(),
                            wbox.right(), wbox.bottom());
    //make_picture_current(fx_win);
    ScrollView::Update();
  }
#endif

  set_global_subloc_code(SUBLOC_NORM);
#if 0
  if (tessedit_print_text) {
    write_cooked_text (word->outword, word->best_choice->string (),
      word->done, done_this_pass, stdout);
  }
#endif
  check_debug_pt (word, 50);
}


/**********************************************************************
 * match_word_pass2
 *
 * Baseline normalize the word and pass it to Tess.
 **********************************************************************/

void match_word_pass2(                 //recog one word
                      WERD_RES *word,  //word to do
                      ROW *row,
                      float x_height) {
  WERD *bln_word;                //baseline norm copy
                                 //detailed results
  BLOB_CHOICE_LIST_CLIST local_blob_choices;
  BLOB_CHOICE_LIST_CLIST *blob_choices;

  if (save_best_choices)
    blob_choices = new BLOB_CHOICE_LIST_CLIST();
  else
    blob_choices = &local_blob_choices;

  set_global_subsubloc_code(SUBSUBLOC_OTHER);
  if (matcher_fp != NULL) {
    word_answer = (char *) word->word->text ();
    if (word_answer != NULL && word_answer[0] == '\0')
      word_answer = NULL;
  }
  matcher_pass = 0;
  bln_word = make_bln_copy (word->word, row, x_height, &word->denorm);
  set_global_subsubloc_code(SUBSUBLOC_TESS);
  if (tessedit_training_tess)
    word->best_choice = correct_segment_pass2 (bln_word,
      &word->denorm,
      tess_default_matcher,
      tess_training_tester,
      word->raw_choice,
      blob_choices, word->outword);
  else if (tessedit_dump_choices)
    word->best_choice = test_segment_pass2 (bln_word,
        &word->denorm,
        tess_default_matcher,
        choice_dump_tester,
        word->raw_choice,
        blob_choices, word->outword);
  //      else if (tessedit_training_wiseowl)
  //              best_choice=correct_segment_pass2( word, &denorm,
  //                                                                                                        tess_default_matcher,wo_learn,
  //                                                                                                        raw_choice,blob_choices,outword);
  //      else if (tessedit_matcher_is_wiseowl)
  //              best_choice=tess_segment_pass2( word, &denorm, wo_classify,
  //                                                                                                raw_choice, blob_choices, outword);
  else {
    word->best_choice = tess_segment_pass2 (bln_word, &word->denorm,
      tess_default_matcher,
      word->raw_choice, blob_choices,
      word->outword);
  }
  set_global_subsubloc_code(SUBSUBLOC_OTHER);
  /*
     Test for TESS screw up on word. Recog_word has already ensured that the
     choice list, outword blob lists and best_choice string are the same
     length. A TESS screw up is indicated by a blank filled or 0 length string.
   */
  if ((word->best_choice->string ().length () == 0) ||
    (strspn (word->best_choice->string ().string (), " ") ==
  word->best_choice->string ().length ())) {
    word->tess_failed = TRUE;
    word->reject_map.initialise (word->best_choice->string ().length ());
    word->reject_map.rej_word_tess_failure ();
    //              tprintf("Empty word produced\n");
  }
  else {
    if ((word->best_choice->lengths ().length () !=
      word->outword->blob_list ()->length ()) ||
    (word->best_choice->lengths ().length () != blob_choices->length ())) {
      tprintf
        ("ASSERT FAIL String:\"%s\"; Strlen=%d; #Blobs=%d; #Choices=%d\n",
        word->best_choice->string ().string (),
        word->best_choice->lengths ().length (),
        word->outword->blob_list ()->length (), blob_choices->length ());
    }
    ASSERT_HOST (word->best_choice->lengths ().length () ==
      word->outword->blob_list ()->length ());
    ASSERT_HOST (word->best_choice->lengths ().length () ==
      blob_choices->length ());

    word->tess_failed = FALSE;
    if (word->word->flag (W_REP_CHAR)) {
      fix_rep_char(word);
    }
    else {
      fix_quotes (word->best_choice,
        word->outword, blob_choices);
      if (tessedit_fix_hyphens)
        fix_hyphens (word->best_choice,
          word->outword, blob_choices);
      /* Dont trust fix_quotes! - though I think I've fixed the bug */
      if ((word->best_choice->lengths ().length () !=
           word->outword->blob_list ()->length ()) ||
          (word->best_choice->lengths ().length () !=
           blob_choices->length ())) {
        #ifndef SECURE_NAMES
        tprintf
          ("POST FIX_QUOTES FAIL String:\"%s\"; Strlen=%d; #Blobs=%d; #Choices=%d\n",
           word->best_choice->string ().string (),
           word->best_choice->lengths ().length (),
           word->outword->blob_list ()->length (),
           blob_choices->length ());
        #endif

      }
      ASSERT_HOST (word->best_choice->lengths ().length () ==
        word->outword->blob_list ()->length ());
      ASSERT_HOST (word->best_choice->lengths ().length () ==
        blob_choices->length ());

      word->tess_accepted = tess_acceptable_word (word->best_choice,
        word->raw_choice);

      make_reject_map (word, blob_choices, row, 2);
    }
  }

  // Save best choices in the WERD_CHOICE if needed
  if (blob_choices != &local_blob_choices)
    word->best_choice->set_blob_choices(blob_choices);
  else
    blob_choices->deep_clear();

  delete bln_word;
  assert (word->raw_choice != NULL);
}


/*************************************************************************
 * fix_rep_char()
 * The word is a repeated char. Find the repeated char character. Make a reject
 * string which rejects any char other than the voted char. Set the word to done
 * to stop rematching it.
 *
 *************************************************************************/
void fix_rep_char(                //Repeated char word
                  WERD_RES *word  //word to do
                 ) {
  struct REP_CH
  {
    char ch[UNICHAR_LEN + 1];
    int count;
  };

  REP_CH *rep_ch;                //array of char counts
  int word_len;
  int rep_ch_count = 0;          //how many unique chs
  const char *word_str;          //the repeated chs
  int i, j;
  int offset;
  int total = 0;
  int max = 0;
  char *maxch = NULL;              //Most common char

  word_str = word->best_choice->string ().string ();
  word_len = word->best_choice->lengths ().length ();;
  rep_ch = (REP_CH *) alloc_mem (word_len * sizeof (REP_CH));
  for (i = 0, offset = 0; i < word_len;
       offset += word->best_choice->lengths()[i++]) {
    for (j = 0; j < rep_ch_count &&
             strncmp(rep_ch[j].ch, word_str + offset,
                     word->best_choice->lengths()[i]) != 0; j++);
    if (j < rep_ch_count)
      rep_ch[j].count++;
    else {
      strncpy(rep_ch[rep_ch_count].ch, word_str + offset,
              word->best_choice->lengths()[i]);
      rep_ch[rep_ch_count].ch[word->best_choice->lengths()[i]] = '\0';
      rep_ch[rep_ch_count].count = 1;
      rep_ch_count++;
    }
  }

  for (j = 0; j < rep_ch_count; j++) {
    total += rep_ch[j].count;
    if ((rep_ch[j].count > max) && (*rep_ch[j].ch != ' ')) {
      max = rep_ch[j].count;
      maxch = rep_ch[j].ch;
    }