tesseract::ParagraphModelSmearer Class Reference

#include <paragraphs_internal.h>

List of all members.

Public Member Functions
	ParagraphModelSmearer (GenericVector< RowScratchRegisters > *rows, int row_start, int row_end, ParagraphTheory *theory)
void	Smear ()

---

Detailed Description

Definition at line 234 of file paragraphs_internal.h.

---

Constructor & Destructor Documentation

tesseract::ParagraphModelSmearer::ParagraphModelSmearer	(	GenericVector< RowScratchRegisters > *	rows,
		int	row_start,
		int	row_end,
		ParagraphTheory *	theory
	)

Definition at line 1258 of file paragraphs.cpp.

        : theory_(theory), rows_(rows), row_start_(row_start),
          row_end_(row_end) {
  if (!AcceptableRowArgs(0, 0, __func__, rows, row_start, row_end)) {
    row_start_ = 0;
    row_end_ = 0;
    return;
  }
  SetOfModels no_models;
  for (int row = row_start - 1; row <= row_end; row++) {
    open_models_.push_back(no_models);
  }
}

---

Member Function Documentation

void tesseract::ParagraphModelSmearer::Smear

(

)

Definition at line 1305 of file paragraphs.cpp.

                                  {
  CalculateOpenModels(row_start_, row_end_);

  // For each row which we're unsure about (that is, it is LT_UNKNOWN or
  // we have multiple LT_START hypotheses), see if there's a model that
  // was recently used (an "open" model) which might model it well.
  for (int i = row_start_; i < row_end_; i++) {
    RowScratchRegisters &row = (*rows_)[i];
    if (row.ri_->num_words == 0)
      continue;

    // Step One:
    //   Figure out if there are "open" models which are left-alined or
    //   right-aligned.  This is important for determining whether the
    //   "first" word in a row would fit at the "end" of the previous row.
    bool left_align_open = false;
    bool right_align_open = false;
    for (int m = 0; m < OpenModels(i).size(); m++) {
      switch (OpenModels(i)[m]->justification()) {
        case JUSTIFICATION_LEFT: left_align_open = true; break;
        case JUSTIFICATION_RIGHT: right_align_open = true; break;
        default: left_align_open = right_align_open = true;
      }
    }
    // Step Two:
    //   Use that knowledge to figure out if this row is likely to
    //   start a paragraph.
    bool likely_start;
    if (i == 0) {
      likely_start = true;
    } else {
      if ((left_align_open && right_align_open) ||
          (!left_align_open && !right_align_open)) {
        likely_start = LikelyParagraphStart((*rows_)[i - 1], row,
                                            JUSTIFICATION_LEFT) ||
                       LikelyParagraphStart((*rows_)[i - 1], row,
                                            JUSTIFICATION_RIGHT);
      } else if (left_align_open) {
        likely_start = LikelyParagraphStart((*rows_)[i - 1], row,
                                            JUSTIFICATION_LEFT);
      } else {
        likely_start = LikelyParagraphStart((*rows_)[i - 1], row,
                                            JUSTIFICATION_RIGHT);
      }
    }

    // Step Three:
    //   If this text line seems like an obvious first line of an
    //   open model, or an obvious continuation of an existing
    //   modelled paragraph, mark it up.
    if (likely_start) {
      // Add Start Hypotheses for all Open models that fit.
      for (int m = 0; m < OpenModels(i).size(); m++) {
        if (ValidFirstLine(rows_, i, OpenModels(i)[m])) {
          row.AddStartLine(OpenModels(i)[m]);
        }
      }
    } else {
      // Add relevant body line hypotheses.
      SetOfModels last_line_models;
      if (i > 0) {
        (*rows_)[i - 1].StrongHypotheses(&last_line_models);
      } else {
        theory_->NonCenteredModels(&last_line_models);
      }
      for (int m = 0; m < last_line_models.size(); m++) {
        const ParagraphModel *model = last_line_models[m];
        if (ValidBodyLine(rows_, i, model))
          row.AddBodyLine(model);
      }
    }

    // Step Four:
    //   If we're still quite unsure about this line, go through all
    //   models in our theory and see if this row could be the start
    //   of any of our  models.
    if (row.GetLineType() == LT_UNKNOWN ||
        (row.GetLineType() == LT_START && !row.UniqueStartHypothesis())) {
      SetOfModels all_models;
      theory_->NonCenteredModels(&all_models);
      for (int m = 0; m < all_models.size(); m++) {
        if (ValidFirstLine(rows_, i, all_models[m])) {
          row.AddStartLine(all_models[m]);
        }
      }
    }
    // Step Five:
    //   Since we may have updated the hypotheses about this row, we need
    //   to recalculate the Open models for the rest of rows[i + 1, row_end)
    if (row.GetLineType() != LT_UNKNOWN) {
      CalculateOpenModels(i + 1, row_end_);
    }
  }
}

---

The documentation for this class was generated from the following files:

• tesseract-ocr/ccmain/paragraphs_internal.h
• tesseract-ocr/ccmain/paragraphs.cpp