C++ TBOX::height方法代码示例

float blob_noise_score(PBLOB *blob) {
  OUTLINE_IT outline_it;
  TBOX box;                       //BB of outline
  inT16 outline_count = 0;
  inT16 max_dimension;
  inT16 largest_outline_dimension = 0;

  outline_it.set_to_list (blob->out_list ());
  for (outline_it.mark_cycle_pt ();
  !outline_it.cycled_list (); outline_it.forward ()) {
    outline_count++;
    box = outline_it.data ()->bounding_box ();
    if (box.height () > box.width ())
      max_dimension = box.height ();
    else
      max_dimension = box.width ();

    if (largest_outline_dimension < max_dimension)
      largest_outline_dimension = max_dimension;
  }

  if (fixsp_noise_score_fixing) {
    if (outline_count > 5)
                                 //penalise LOTS of blobs
      largest_outline_dimension *= 2;

    box = blob->bounding_box ();

    if ((box.bottom () > bln_baseline_offset * 4) ||
      (box.top () < bln_baseline_offset / 2))
                                 //Lax blob is if high or low
      largest_outline_dimension /= 2;
  }
  return largest_outline_dimension;
}

float Tesseract::blob_noise_score(TBLOB *blob) {
  TBOX box;                       // BB of outline
  inT16 outline_count = 0;
  inT16 max_dimension;
  inT16 largest_outline_dimension = 0;

  for (TESSLINE* ol = blob->outlines; ol != NULL; ol= ol->next) {
    outline_count++;
    box = ol->bounding_box();
    if (box.height() > box.width()) {
      max_dimension = box.height();
    } else {
      max_dimension = box.width();
    }

    if (largest_outline_dimension < max_dimension)
      largest_outline_dimension = max_dimension;
  }

  if (outline_count > 5) {
    // penalise LOTS of blobs
    largest_outline_dimension *= 2;
  }

  box = blob->bounding_box();
  if (box.bottom() > kBlnBaselineOffset * 4 ||
      box.top() < kBlnBaselineOffset / 2) {
    // Lax blob is if high or low
    largest_outline_dimension /= 2;
  }

  return largest_outline_dimension;
}

// Tests each blob in the list to see if it is certain non-text using 2
// conditions:
// 1. blob overlaps a cell with high value in noise_density_ (previously set
// by ComputeNoiseDensity).
// OR 2. The blob overlaps more than max_blob_overlaps in *this grid. This
// condition is disabled with max_blob_overlaps == -1.
// If it does, the blob is declared non-text, and is used to mark up the
// nontext_mask. Such blobs are fully deleted, and non-noise blobs have their
// neighbours reset, as they may now point to deleted data.
// WARNING: The blobs list blobs may be in the *this grid, but they are
// not removed. If any deleted blobs might be in *this, then this must be
// Clear()ed immediately after MarkAndDeleteNonTextBlobs is called.
// If the win is not NULL, deleted blobs are drawn on it in red, and kept
// blobs are drawn on it in ok_color.
void CCNonTextDetect::MarkAndDeleteNonTextBlobs(BLOBNBOX_LIST* blobs,
        int max_blob_overlaps,
        ScrollView* win,
        ScrollView::Color ok_color,
        Pix* nontext_mask) {
    int imageheight = tright().y() - bleft().x();
    BLOBNBOX_IT blob_it(blobs);
    BLOBNBOX_LIST dead_blobs;
    BLOBNBOX_IT dead_it(&dead_blobs);
    for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
        BLOBNBOX* blob = blob_it.data();
        TBOX box = blob->bounding_box();
        if (!noise_density_->RectMostlyOverThreshold(box, max_noise_count_) &&
                (max_blob_overlaps < 0 ||
                 !BlobOverlapsTooMuch(blob, max_blob_overlaps))) {
            blob->ClearNeighbours();
#ifndef GRAPHICS_DISABLED
            if (win != NULL)
                blob->plot(win, ok_color, ok_color);
#endif  // GRAPHICS_DISABLED
        } else {
            if (noise_density_->AnyZeroInRect(box)) {
                // There is a danger that the bounding box may overlap real text, so
                // we need to render the outline.
                Pix* blob_pix = blob->cblob()->render_outline();
                pixRasterop(nontext_mask, box.left(), imageheight - box.top(),
                            box.width(), box.height(), PIX_SRC | PIX_DST,
                            blob_pix, 0, 0);
                pixDestroy(&blob_pix);
            } else {
                if (box.area() < gridsize() * gridsize()) {
                    // It is a really bad idea to make lots of small components in the
                    // photo mask, so try to join it to a bigger area by expanding the
                    // box in a way that does not touch any zero noise density cell.
                    box = AttemptBoxExpansion(box, *noise_density_, gridsize());
                }
                // All overlapped cells are non-zero, so just mark the rectangle.
                pixRasterop(nontext_mask, box.left(), imageheight - box.top(),
                            box.width(), box.height(), PIX_SET, NULL, 0, 0);
            }
#ifndef GRAPHICS_DISABLED
            if (win != NULL)
                blob->plot(win, ScrollView::RED, ScrollView::RED);
#endif  // GRAPHICS_DISABLED
            // It is safe to delete the cblob now, as it isn't used by the grid
            // or BlobOverlapsTooMuch, and the BLOBNBOXes will go away with the
            // dead_blobs list.
            // TODO(rays) delete the delete when the BLOBNBOX destructor deletes
            // the cblob.
            delete blob->cblob();
            dead_it.add_to_end(blob_it.extract());
        }
    }
}

PBLOB::PBLOB(                //constructor
             C_BLOB *cblob,  //compact blob
             float xheight   //height of line
            ) {
  TBOX bbox;                      //bounding box

  if (!cblob->out_list ()->empty ()) {
                                 //get bounding box
    bbox = cblob->bounding_box ();
    if (bbox.height () > xheight)
      xheight = bbox.height ();  //max of line and blob
                                 //copy it
    approximate_outline_list (cblob->out_list (), &outlines, xheight);
  }
}

// Compute the distance from the from_box to the to_box using curved
// projection space. Separation that involves a decrease in projection
// density (moving from the from_box to the to_box) is weighted more heavily
// than constant density, and an increase is weighted less.
// If horizontal_textline is true, then curved space is used vertically,
// as for a diacritic on the edge of a textline.
// The projection uses original image coords, so denorm is used to get
// back to the image coords from box/part space.
// How the calculation works: Think of a diacritic near a textline.
// Distance is measured from the far side of the from_box to the near side of
// the to_box. Shown is the horizontal textline case.
//          |------^-----|
//          | from | box |
//          |------|-----|
//   perpendicular |
//          <------v-------->|--------------------|
//                  parallel |     to box         |
//                           |--------------------|
// Perpendicular distance uses "curved space" See VerticalDistance below.
// Parallel distance is linear.
// Result is perpendicular_gap + parallel_gap / kParaPerpDistRatio.
int TextlineProjection::DistanceOfBoxFromBox(const TBOX& from_box,
                                             const TBOX& to_box,
                                             bool horizontal_textline,
                                             const DENORM* denorm,
                                             bool debug) const {
  // The parallel_gap is the horizontal gap between a horizontal textline and
  // the box. Analogous for vertical.
  int parallel_gap = 0;
  // start_pt is the box end of the line to be modified for curved space.
  TPOINT start_pt;
  // end_pt is the partition end of the line to be modified for curved space.
  TPOINT end_pt;
  if (horizontal_textline) {
    parallel_gap = from_box.x_gap(to_box) + from_box.width();
    start_pt.x = (from_box.left() + from_box.right()) / 2;
    end_pt.x = start_pt.x;
    if (from_box.top() - to_box.top() >= to_box.bottom() - from_box.bottom()) {
      start_pt.y = from_box.top();
      end_pt.y = MIN(to_box.top(), start_pt.y);
    } else {
      start_pt.y = from_box.bottom();
      end_pt.y = MAX(to_box.bottom(), start_pt.y);
    }
  } else {
    parallel_gap = from_box.y_gap(to_box) + from_box.height();
    if (from_box.right() - to_box.right() >= to_box.left() - from_box.left()) {
      start_pt.x = from_box.right();
      end_pt.x = MIN(to_box.right(), start_pt.x);
    } else {
      start_pt.x = from_box.left();
      end_pt.x = MAX(to_box.left(), start_pt.x);
    }
    start_pt.y = (from_box.bottom() + from_box.top()) / 2;
    end_pt.y = start_pt.y;
  }
  // The perpendicular gap is the max vertical distance gap out of:
  // top of from_box to to_box top and bottom of from_box to to_box bottom.
  // This value is then modified for curved projection space.
  // Analogous for vertical.
  int perpendicular_gap = 0;
  // If start_pt == end_pt, then the from_box lies entirely within the to_box
  // (in the perpendicular direction), so we don't need to calculate the
  // perpendicular_gap.
  if (start_pt.x != end_pt.x || start_pt.y != end_pt.y) {
    if (denorm != NULL) {
      // Denormalize the start and end.
      denorm->DenormTransform(NULL, start_pt, &start_pt);
      denorm->DenormTransform(NULL, end_pt, &end_pt);
    }
    if (abs(start_pt.y - end_pt.y) >= abs(start_pt.x - end_pt.x)) {
      perpendicular_gap = VerticalDistance(debug, start_pt.x, start_pt.y,
                                           end_pt.y);
    } else {
      perpendicular_gap = HorizontalDistance(debug, start_pt.x, end_pt.x,
                                             start_pt.y);
    }
  }
  // The parallel_gap weighs less than the perpendicular_gap.
  return perpendicular_gap + parallel_gap / kParaPerpDistRatio;
}

// Sets up the DENORM to execute a non-linear transformation based on
// preserving an even distribution of stroke edges. The transformation
// operates only within the given box.
// x_coords is a collection of the x-coords of vertical edges for each
// y-coord starting at box.bottom().
// y_coords is a collection of the y-coords of horizontal edges for each
// x-coord starting at box.left().
// Eg x_coords[0] is a collection of the x-coords of edges at y=bottom.
// Eg x_coords[1] is a collection of the x-coords of edges at y=bottom + 1.
// The second-level vectors must all be sorted in ascending order.
// See comments on the helper functions above for more details.
void DENORM::SetupNonLinear(
    const DENORM* predecessor, const TBOX& box, float target_width,
    float target_height, float final_xshift, float final_yshift,
    const GenericVector<GenericVector<int> >& x_coords,
    const GenericVector<GenericVector<int> >& y_coords) {
  Clear();
  predecessor_ = predecessor;
  // x_map_ and y_map_ store a mapping from input x and y coordinate to output
  // x and y coordinate, based on scaling to the supplied target_width and
  // target_height.
  x_map_ = new GenericVector<float>;
  y_map_ = new GenericVector<float>;
  // Set a 2-d image array to the run lengths at each pixel.
  int width = box.width();
  int height = box.height();
  GENERIC_2D_ARRAY<int> minruns(width, height, 0);
  ComputeRunlengthImage(box, x_coords, y_coords, &minruns);
  // Edge density is the sum of the inverses of the run lengths. Compute
  // edge density projection profiles.
  ComputeEdgeDensityProfiles(box, minruns, x_map_, y_map_);
  // Convert the edge density profiles to the coordinates by multiplying by
  // the desired size and accumulating.
  (*x_map_)[width] = target_width;
  for (int x = width - 1; x >= 0; --x) {
    (*x_map_)[x] = (*x_map_)[x + 1] - (*x_map_)[x] * target_width;
  }
  (*y_map_)[height] = target_height;
  for (int y = height - 1; y >= 0; --y) {
    (*y_map_)[y] = (*y_map_)[y + 1] - (*y_map_)[y] * target_height;
  }
  x_origin_ = box.left();
  y_origin_ = box.bottom();
  final_xshift_ = final_xshift;
  final_yshift_ = final_yshift;
}

// Converts the run-length image (see above to the edge density profiles used
// for scaling, thus:
//  ______________
// |7 1_1_1_1_1 7|  = 5.28
// |1|5 5 1 5 5|1|  = 3.8
// |1|2 2|1|2 2|1|  = 5
// |1|2 2|1|2 2|1|  = 5
// |1|2 2|1|2 2|1|  = 5
// |1|2 2|1|2 2|1|  = 5
// |1|5_5_1_5_5|1|  = 3.8
// |7_1_1_1_1_1_7|  = 5.28
//  6 4 4 8 4 4 6
//  . . . . . . .
//  2 4 4 0 4 4 2
//  8           8
// Each profile is the sum of the reciprocals of the pixels in the image in
// the appropriate row or column, and these are then normalized to sum to 1.
// On output hx, hy contain an extra element, which will eventually be used
// to guarantee that the top/right edge of the box (and anything beyond) always
// gets mapped to the maximum target coordinate.
static void ComputeEdgeDensityProfiles(const TBOX& box,
                                       const GENERIC_2D_ARRAY<int>& minruns,
                                       GenericVector<float>* hx,
                                       GenericVector<float>* hy) {
  int width = box.width();
  int height = box.height();
  hx->init_to_size(width + 1, 0.0);
  hy->init_to_size(height + 1, 0.0);
  double total = 0.0;
  for (int iy = 0; iy < height; ++iy) {
    for (int ix = 0; ix < width; ++ix) {
      int run = minruns(ix, iy);
      if (run == 0) run = 1;
      float density = 1.0f / run;
      (*hx)[ix] += density;
      (*hy)[iy] += density;
    }
    total += (*hy)[iy];
  }
  // Normalize each profile to sum to 1.
  if (total > 0.0) {
    for (int ix = 0; ix < width; ++ix) {
      (*hx)[ix] /= total;
    }
    for (int iy = 0; iy < height; ++iy) {
      (*hy)[iy] /= total;
    }
  }
  // There is an extra element in each array, so initialize to 1.
  (*hx)[width] = 1.0f;
  (*hy)[height] = 1.0f;
}

// Adds edges to the given vectors.
// For all the edge steps in all the outlines, or polygonal approximation
// where there are no edge steps, collects the steps into x_coords/y_coords.
// x_coords is a collection of the x-coords of vertical edges for each
// y-coord starting at box.bottom().
// y_coords is a collection of the y-coords of horizontal edges for each
// x-coord starting at box.left().
// Eg x_coords[0] is a collection of the x-coords of edges at y=bottom.
// Eg x_coords[1] is a collection of the x-coords of edges at y=bottom + 1.
void TBLOB::GetEdgeCoords(const TBOX& box,
                          GenericVector<GenericVector<int> >* x_coords,
                          GenericVector<GenericVector<int> >* y_coords) const {
  GenericVector<int> empty;
  x_coords->init_to_size(box.height(), empty);
  y_coords->init_to_size(box.width(), empty);
  CollectEdges(box, nullptr, nullptr, x_coords, y_coords);
  // Sort the output vectors.
  for (int i = 0; i < x_coords->size(); ++i) (*x_coords)[i].sort();
  for (int i = 0; i < y_coords->size(); ++i) (*y_coords)[i].sort();
}

/**********************************************************************
 * char_box_to_tbox
 *
 * Create a TBOX from a character bounding box. If nonzero, the
 * x_offset accounts for any additional padding of the word box that
 * should be taken into account.
 *
 **********************************************************************/
TBOX char_box_to_tbox(Box* char_box, TBOX word_box, int x_offset) {
  l_int32 left;
  l_int32 top;
  l_int32 width;
  l_int32 height;
  l_int32 right;
  l_int32 bottom;

  boxGetGeometry(char_box, &left, &top, &width, &height);
  left += word_box.left() - x_offset;
  right = left + width;
  top = word_box.bottom() + word_box.height() - top;
  bottom = top - height;
  return TBOX(left, bottom, right, top);
}

// Helper for SetupNonLinear computes an image of shortest run-lengths from
// the x/y edges provided.
// Based on "A nonlinear normalization method for handprinted Kanji character
// recognition -- line density equalization" by Hiromitsu Yamada et al.
// Eg below is an O in a 1-pixel margin-ed bounding box and the corresponding
//  ______________     input x_coords and y_coords.
// |  _________  |     <empty>
// | |    _    | |     1, 6
// | |   | |   | |     1, 3, 4, 6
// | |   | |   | |     1, 3, 4, 6
// | |   | |   | |     1, 3, 4, 6
// | |   |_|   | |     1, 3, 4, 6
// | |_________| |     1, 6
// |_____________|     <empty>
//  E 1 1 1 1 1 E
//  m 7 7 2 7 7 m
//  p     6     p
//  t     7     t
//  y           y
// The output image contains the min of the x and y run-length (distance
// between edges) at each coordinate in the image thus:
//  ______________
// |7 1_1_1_1_1 7|
// |1|5 5 1 5 5|1|
// |1|2 2|1|2 2|1|
// |1|2 2|1|2 2|1|
// |1|2 2|1|2 2|1|
// |1|2 2|1|2 2|1|
// |1|5_5_1_5_5|1|
// |7_1_1_1_1_1_7|
// Note that the input coords are all integer, so all partial pixels are dealt
// with elsewhere. Although it is nice for outlines to be properly connected
// and continuous, there is no requirement that they be as such, so they could
// have been derived from a flaky source, such as greyscale.
// This function works only within the provided box, and it is assumed that the
// input x_coords and y_coords have already been translated to have the bottom-
// left of box as the origin. Although an output, the minruns should have been
// pre-initialized to be the same size as box. Each element will contain the
// minimum of x and y run-length as shown above.
static void ComputeRunlengthImage(
    const TBOX& box,
    const GenericVector<GenericVector<int> >& x_coords,
    const GenericVector<GenericVector<int> >& y_coords,
    GENERIC_2D_ARRAY<int>* minruns) {
  int width = box.width();
  int height = box.height();
  ASSERT_HOST(minruns->dim1() == width);
  ASSERT_HOST(minruns->dim2() == height);
  // Set a 2-d image array to the run lengths at each pixel.
  for (int ix = 0; ix < width; ++ix) {
    int y = 0;
    for (int i = 0; i < y_coords[ix].size(); ++i) {
      int y_edge = ClipToRange(y_coords[ix][i], 0, height);
      int gap = y_edge - y;
      // Every pixel between the last and current edge get set to the gap.
      while (y < y_edge) {
        (*minruns)(ix, y) = gap;
        ++y;
      }
    }
    // Pretend there is a bounding box of edges all around the image.
    int gap = height - y;
    while (y < height) {
      (*minruns)(ix, y) = gap;
      ++y;
    }
  }
  // Now set the image pixels the the MIN of the x and y runlengths.
  for (int iy = 0; iy < height; ++iy) {
    int x = 0;
    for (int i = 0; i < x_coords[iy].size(); ++i) {
      int x_edge = ClipToRange(x_coords[iy][i], 0, width);
      int gap = x_edge - x;
      while (x < x_edge) {
        if (gap < (*minruns)(x, iy))
          (*minruns)(x, iy) = gap;
        ++x;
      }
    }
    int gap = width - x;
    while (x < width) {
      if (gap < (*minruns)(x, iy))
        (*minruns)(x, iy) = gap;
      ++x;
    }
  }
}

// Given an input pix, and a box, the sides of the box are shrunk inwards until
// they bound any black pixels found within the original box.
// The function converts between tesseract coords and the pix coords assuming
// that this pix is full resolution equal in size to the original image.
// Returns an empty box if there are no black pixels in the source box.
static TBOX BoundsWithinBox(Pix* pix, const TBOX& box) {
  int im_height = pixGetHeight(pix);
  Box* input_box = boxCreate(box.left(), im_height - box.top(),
                             box.width(), box.height());
  Box* output_box = NULL;
  pixClipBoxToForeground(pix, input_box, NULL, &output_box);
  TBOX result_box;
  if (output_box != NULL) {
    l_int32 x, y, width, height;
    boxGetGeometry(output_box, &x, &y, &width, &height);
    result_box.set_left(x);
    result_box.set_right(x + width);
    result_box.set_top(im_height - y);
    result_box.set_bottom(result_box.top() - height);
    boxDestroy(&output_box);
  }
  boxDestroy(&input_box);
  return result_box;
}

BOOL8 suspect_fullstop(WERD_RES *word, inT16 i) {
  float aspect_ratio;
  PBLOB_LIST *blobs = word->outword->blob_list ();
  PBLOB_IT blob_it(blobs);
  inT16 j;
  TBOX box;
  inT16 width;
  inT16 height;

  for (j = 0; j < i; j++)
    blob_it.forward ();

  box = blob_it.data ()->bounding_box ();

  width = box.width ();
  height = box.height ();

  aspect_ratio = ((width > height) ? ((float) width) / height :
  ((float) height) / width);

  return (aspect_ratio > tessed_fullstop_aspect_ratio);
}

/**********************************************************************
 * make_rotated_tess_blob
 *
 * Make a single Tess style blob, applying the given rotation and
 * renormalizing.
 **********************************************************************/
TBLOB *make_rotated_tess_blob(const DENORM* denorm, PBLOB *blob,
                              BOOL8 flatten) {
  if (denorm != NULL && denorm->block() != NULL &&
      denorm->block()->classify_rotation().y() != 0.0) {
    TBOX box = blob->bounding_box();
    int src_width = box.width();
    int src_height = box.height();
    src_width = static_cast<int>(src_width / denorm->scale() + 0.5);
    src_height = static_cast<int>(src_height / denorm->scale() + 0.5);
    int x_middle = (box.left() + box.right()) / 2;
    int y_middle = (box.top() + box.bottom()) / 2;
    PBLOB* rotated_blob = PBLOB::deep_copy(blob);
    rotated_blob->move(FCOORD(-x_middle, -y_middle));
    rotated_blob->rotate(denorm->block()->classify_rotation());
    ICOORD median_size = denorm->block()->median_size();
    int tolerance = median_size.x() / 8;
    // TODO(dsl/rays) find a better normalization solution. In the mean time
    // make it work for CJK by normalizing for Cap height in the same way
    // as is applied in compute_block_xheight when the row is presumed to
    // be ALLCAPS, i.e. the x-height is the fixed fraction
    // blob height * textord_merge_x / (textord_merge_x + textord_merge_asc)
    if (NearlyEqual(src_width, static_cast<int>(median_size.x()), tolerance) &&
        NearlyEqual(src_height, static_cast<int>(median_size.y()), tolerance)) {
      float target_height = bln_x_height * (textord_merge_x + textord_merge_asc)
                          / textord_merge_x;
      rotated_blob->scale(target_height / box.width());
      rotated_blob->move(FCOORD(0.0f,
                                bln_baseline_offset -
                                  rotated_blob->bounding_box().bottom()));
    }
    TBLOB* result = make_tess_blob(rotated_blob, flatten);
    delete rotated_blob;
    return result;
  } else {
    return make_tess_blob(blob, flatten);
  }
}

// Fixes the block so it obeys all the rules:
// Must have at least one ROW.
// Must have at least one WERD.
// WERDs contain a fake blob.
void Textord::cleanup_nontext_block(BLOCK* block) {
  // Non-text blocks must contain at least one row.
  ROW_IT row_it(block->row_list());
  if (row_it.empty()) {
    TBOX box = block->bounding_box();
    float height = box.height();
    inT32 xstarts[2] = {box.left(), box.right()};
    double coeffs[3] = {0.0, 0.0, static_cast<double>(box.bottom())};
    ROW* row = new ROW(1, xstarts, coeffs, height / 2.0f, height / 4.0f,
                       height / 4.0f, 0, 1);
    row_it.add_after_then_move(row);
  }
  // Each row must contain at least one word.
  for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
    ROW* row = row_it.data();
    WERD_IT w_it(row->word_list());
    if (w_it.empty()) {
      // Make a fake blob to put in the word.
      TBOX box = block->row_list()->singleton() ? block->bounding_box()
                                                : row->bounding_box();
      C_BLOB* blob = C_BLOB::FakeBlob(box);
      C_BLOB_LIST blobs;
      C_BLOB_IT blob_it(&blobs);
      blob_it.add_after_then_move(blob);
      WERD* word = new WERD(&blobs, 0, NULL);
      w_it.add_after_then_move(word);
    }
    // Each word must contain a fake blob.
    for (w_it.mark_cycle_pt(); !w_it.cycled_list(); w_it.forward()) {
      WERD* word = w_it.data();
      // Just assert that this is true, as it would be useful to find
      // out why it isn't.
      ASSERT_HOST(!word->cblob_list()->empty());
    }
    row->recalc_bounding_box();
  }
}

// Returns true if the blob is small enough to be a large speckle.
bool Classify::LargeSpeckle(const TBLOB &blob) {
  double speckle_size = kBlnXHeight * speckle_large_max_size;
  TBOX bbox = blob.bounding_box();
  return bbox.width() < speckle_size && bbox.height() < speckle_size;
}