C++ TBOX类代码示例

// 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;
}

// Evaluates the textlineiness of a ColPartition. Uses EvaluateBox below,
// but uses the median top/bottom for horizontal and median left/right for
// vertical instead of the bounding box edges.
// Evaluates for both horizontal and vertical and returns the best result,
// with a positive value for horizontal and a negative value for vertical.
int TextlineProjection::EvaluateColPartition(const ColPartition& part,
                                             const DENORM* denorm,
                                             bool debug) const {
  if (part.IsSingleton())
    return EvaluateBox(part.bounding_box(), denorm, debug);
  // Test vertical orientation.
  TBOX box = part.bounding_box();
  // Use the partition median for left/right.
  box.set_left(part.median_left());
  box.set_right(part.median_right());
  int vresult = EvaluateBox(box, denorm, debug);

  // Test horizontal orientation.
  box = part.bounding_box();
  // Use the partition median for top/bottom.
  box.set_top(part.median_top());
  box.set_bottom(part.median_bottom());
  int hresult = EvaluateBox(box, denorm, debug);
  if (debug) {
    tprintf("Partition hresult=%d, vresult=%d from:", hresult, vresult);
    part.bounding_box().print();
    part.Print();
  }
  return hresult >= -vresult ? hresult : vresult;
}

// Swaps the outlines of *this and next if needed to keep the centers in
// increasing x.
void TBLOB::CorrectBlobOrder(TBLOB* next) {
  TBOX box = bounding_box();
  TBOX next_box = next->bounding_box();
  if (box.x_middle() > next_box.x_middle()) {
    Swap(&outlines, &next->outlines);
  }
}

static void PrintBoxWidths(BLOBNBOX* neighbour) {
  TBOX nbox = neighbour->bounding_box();
  tprintf("Box (%d,%d)->(%d,%d): h-width=%.1f, v-width=%.1f p-width=%1.f\n",
          nbox.left(), nbox.bottom(), nbox.right(), nbox.top(),
          neighbour->horz_stroke_width(), neighbour->vert_stroke_width(),
          2.0 * neighbour->cblob()->area()/neighbour->cblob()->perimeter());
}

void OL_BUCKETS::extract_children(                     // recursive count
                                  C_OUTLINE *outline,  // parent outline
                                  C_OUTLINE_IT *it     // destination iterator
                                 ) {
  inT16 xmin, xmax;              // coord limits
  inT16 ymin, ymax;
  inT16 xindex, yindex;          // current bucket
  TBOX olbox;
  C_OUTLINE_IT child_it;         // search iterator

  olbox = outline->bounding_box();
  xmin =(olbox.left() - bl.x()) / BUCKETSIZE;
  xmax =(olbox.right() - bl.x()) / BUCKETSIZE;
  ymin =(olbox.bottom() - bl.y()) / BUCKETSIZE;
  ymax =(olbox.top() - bl.y()) / BUCKETSIZE;
  for (yindex = ymin; yindex <= ymax; yindex++) {
    for (xindex = xmin; xindex <= xmax; xindex++) {
      child_it.set_to_list(&buckets[yindex * bxdim + xindex]);
      for (child_it.mark_cycle_pt(); !child_it.cycled_list();
           child_it.forward()) {
        if (*child_it.data() < *outline) {
          it->add_after_then_move(child_it.extract());
        }
      }
    }
  }
}

// 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;
}

/**********************************************************************
 * render_segmentation
 *
 * Create a list of line segments that represent the list of chunks
 * using the correct segmentation that was supplied as input.
 **********************************************************************/
void render_segmentation(ScrollView *window,
                         TBLOB *chunks,
                         SEARCH_STATE segmentation) {
  TBLOB *blob;
  C_COL color = Black;
  int char_num = -1;
  int chunks_left = 0;

  TBOX bbox;
  if (chunks) bbox = chunks->bounding_box();

  for (blob = chunks; blob != NULL; blob = blob->next) {
    bbox += blob->bounding_box();
    if (chunks_left-- == 0) {
      color = color_list[++char_num % NUM_COLORS];

      if (char_num < segmentation[0])
        chunks_left = segmentation[char_num + 1];
      else
        chunks_left = MAX_INT32;
    }
    render_outline(window, blob->outlines, color);
  }
  window->ZoomToRectangle(bbox.left(), bbox.top(),
                          bbox.right(), bbox.bottom());
}

// Returns an Imagedata containing the image of the given box,
// and ground truth boxes/truth text if available in the input.
// The image is not normalized in any way.
ImageData* Tesseract::GetLineData(const TBOX& line_box,
                                  const GenericVector<TBOX>& boxes,
                                  const GenericVector<STRING>& texts,
                                  int start_box, int end_box,
                                  const BLOCK& block) {
  TBOX revised_box;
  ImageData* image_data = GetRectImage(line_box, block, kImagePadding,
                                       &revised_box);
  if (image_data == NULL) return NULL;
  image_data->set_page_number(applybox_page);
  // Copy the boxes and shift them so they are relative to the image.
  FCOORD block_rotation(block.re_rotation().x(), -block.re_rotation().y());
  ICOORD shift = -revised_box.botleft();
  GenericVector<TBOX> line_boxes;
  GenericVector<STRING> line_texts;
  for (int b = start_box; b < end_box; ++b) {
    TBOX box = boxes[b];
    box.rotate(block_rotation);
    box.move(shift);
    line_boxes.push_back(box);
    line_texts.push_back(texts[b]);
  }
  GenericVector<int> page_numbers;
  page_numbers.init_to_size(line_boxes.size(), applybox_page);
  image_data->AddBoxes(line_boxes, line_texts, page_numbers);
  return image_data;
}

void fixspace_dbg(WERD_RES *word) {
  TBOX box = word->word->bounding_box ();
  BOOL8 show_map_detail = FALSE;
  inT16 i;

  box.print ();
  #ifndef SECURE_NAMES
  tprintf (" \"%s\" ", word->best_choice->string ().string ());
  tprintf ("Blob count: %d (word); %d/%d (outword)\n",
    word->word->gblob_list ()->length (),
    word->outword->gblob_list ()->length (),
    word->outword->rej_blob_list ()->length ());
  word->reject_map.print (debug_fp);
  tprintf ("\n");
  if (show_map_detail) {
    tprintf ("\"%s\"\n", word->best_choice->string ().string ());
    for (i = 0; word->best_choice->string ()[i] != '\0'; i++) {
      tprintf ("**** \"%c\" ****\n", word->best_choice->string ()[i]);
      word->reject_map[i].full_print (debug_fp);
    }
  }

  tprintf ("Tess Accepted: %s\n", word->tess_accepted ? "TRUE" : "FALSE");
  tprintf ("Done flag: %s\n\n", word->done ? "TRUE" : "FALSE");
  #endif
}

// Find a set of blobs that are aligned in the given vertical
// direction with the given blob. Returns a list of aligned
// blobs and the number in the list.
// For other parameters see FindAlignedBlob below.
int AlignedBlob::AlignTabs(const AlignedBlobParams& params,
                           bool top_to_bottom, BLOBNBOX* bbox,
                           BLOBNBOX_CLIST* good_points, int* end_y) {
  int ptcount = 0;
  BLOBNBOX_C_IT it(good_points);

  TBOX box = bbox->bounding_box();
  int x_start = params.right_tab ? box.right() : box.left();
  while (bbox != NULL) {
    // Add the blob to the list if the appropriate side is a tab candidate,
    // or if we are working on a ragged tab.
    if (((params.right_tab && bbox->right_tab_type() != TT_NONE) ||
         (!params.right_tab && bbox->left_tab_type() != TT_NONE) ||
         params.ragged) &&
        (it.empty() || it.data() != bbox)) {
      if (top_to_bottom)
        it.add_before_then_move(bbox);
      else
        it.add_after_then_move(bbox);
      ++ptcount;
    }
    // Find the next blob that is aligned with the current one.
    // FindAlignedBlob guarantees that forward progress will be made in the
    // top_to_bottom direction, and therefore eventually it will return NULL,
    // making this while (bbox != NULL) loop safe.
    bbox = FindAlignedBlob(params, top_to_bottom, bbox, x_start, end_y);
    if (bbox != NULL) {
      box = bbox->bounding_box();
      if (!params.ragged)
        x_start = params.right_tab ? box.right() : box.left();
    }
  }
  return ptcount;
}

/** Handles a click event in a display window. */
void StrokeWidth::HandleClick(int x, int y) {
  BBGrid<BLOBNBOX, BLOBNBOX_CLIST, BLOBNBOX_C_IT>::HandleClick(x, y);
  // Run a radial search for blobs that overlap.
  BlobGridSearch radsearch(this);
  radsearch.StartRadSearch(x, y, 1);
  BLOBNBOX* neighbour;
  FCOORD click(static_cast<float>(x), static_cast<float>(y));
  while ((neighbour = radsearch.NextRadSearch()) != NULL) {
    TBOX nbox = neighbour->bounding_box();
    if (nbox.contains(click) && neighbour->cblob() != NULL) {
      PrintBoxWidths(neighbour);
      if (neighbour->neighbour(BND_LEFT) != NULL)
        PrintBoxWidths(neighbour->neighbour(BND_LEFT));
      if (neighbour->neighbour(BND_RIGHT) != NULL)
        PrintBoxWidths(neighbour->neighbour(BND_RIGHT));
      if (neighbour->neighbour(BND_ABOVE) != NULL)
        PrintBoxWidths(neighbour->neighbour(BND_ABOVE));
      if (neighbour->neighbour(BND_BELOW) != NULL)
        PrintBoxWidths(neighbour->neighbour(BND_BELOW));
      int gaps[BND_COUNT];
      neighbour->NeighbourGaps(gaps);
      tprintf("Left gap=%d, right=%d, above=%d, below=%d, horz=%d, vert=%d\n"
              "Good=    %d        %d        %d        %d\n",
              gaps[BND_LEFT], gaps[BND_RIGHT],
              gaps[BND_ABOVE], gaps[BND_BELOW],
              neighbour->horz_possible(),
              neighbour->vert_possible(),
              neighbour->good_stroke_neighbour(BND_LEFT),
              neighbour->good_stroke_neighbour(BND_RIGHT),
              neighbour->good_stroke_neighbour(BND_ABOVE),
              neighbour->good_stroke_neighbour(BND_BELOW));
      break;
    }
  }
}

// Generates a TrainingSample from a TBLOB. Extracts features and sets
// the bounding box, so classifiers that operate on the image can work.
// TODO(rays) Make BlobToTrainingSample a member of Classify now that
// the FlexFx and FeatureDescription code have been removed and LearnBlob
// is now a member of Classify.
TrainingSample* BlobToTrainingSample(
    const TBLOB& blob, bool nonlinear_norm, INT_FX_RESULT_STRUCT* fx_info,
    GenericVector<INT_FEATURE_STRUCT>* bl_features) {
  GenericVector<INT_FEATURE_STRUCT> cn_features;
  Classify::ExtractFeatures(blob, nonlinear_norm, bl_features,
                            &cn_features, fx_info, nullptr);
  // TODO(rays) Use blob->PreciseBoundingBox() instead.
  TBOX box = blob.bounding_box();
  TrainingSample* sample = nullptr;
  int num_features = fx_info->NumCN;
  if (num_features > 0) {
    sample = TrainingSample::CopyFromFeatures(*fx_info, box, &cn_features[0],
                                              num_features);
  }
  if (sample != nullptr) {
    // Set the bounding box (in original image coordinates) in the sample.
    TPOINT topleft, botright;
    topleft.x = box.left();
    topleft.y = box.top();
    botright.x = box.right();
    botright.y = box.bottom();
    TPOINT original_topleft, original_botright;
    blob.denorm().DenormTransform(nullptr, topleft, &original_topleft);
    blob.denorm().DenormTransform(nullptr, botright, &original_botright);
    sample->set_bounding_box(TBOX(original_topleft.x, original_botright.y,
                                  original_botright.x, original_topleft.y));
  }
  return sample;
}

// Computes the DENORMS for bl(baseline) and cn(character) normalization
// during feature extraction. The input denorm describes the current state
// of the blob, which is usually a baseline-normalized word.
// The Transforms setup are as follows:
// Baseline Normalized (bl) Output:
//   We center the grapheme by aligning the x-coordinate of its centroid with
//   x=128 and leaving the already-baseline-normalized y as-is.
//
// Character Normalized (cn) Output:
//   We align the grapheme's centroid at the origin and scale it
//   asymmetrically in x and y so that the 2nd moments are a standard value
//   (51.2) ie the result is vaguely square.
// If classify_nonlinear_norm is true:
//   A non-linear normalization is setup that attempts to evenly distribute
//   edges across x and y.
//
// Some of the fields of fx_info are also setup:
// Length: Total length of outline.
// Rx:     Rounded y second moment. (Reversed by convention.)
// Ry:     rounded x second moment.
// Xmean:  Rounded x center of mass of the blob.
// Ymean:  Rounded y center of mass of the blob.
void Classify::SetupBLCNDenorms(const TBLOB& blob, bool nonlinear_norm,
                                DENORM* bl_denorm, DENORM* cn_denorm,
                                INT_FX_RESULT_STRUCT* fx_info) {
  // Compute 1st and 2nd moments of the original outline.
  FCOORD center, second_moments;
  int length = blob.ComputeMoments(&center, &second_moments);
  if (fx_info != nullptr) {
    fx_info->Length = length;
    fx_info->Rx = IntCastRounded(second_moments.y());
    fx_info->Ry = IntCastRounded(second_moments.x());

    fx_info->Xmean = IntCastRounded(center.x());
    fx_info->Ymean = IntCastRounded(center.y());
  }
  // Setup the denorm for Baseline normalization.
  bl_denorm->SetupNormalization(nullptr, nullptr, &blob.denorm(), center.x(), 128.0f,
                                1.0f, 1.0f, 128.0f, 128.0f);
  // Setup the denorm for character normalization.
  if (nonlinear_norm) {
    GenericVector<GenericVector<int> > x_coords;
    GenericVector<GenericVector<int> > y_coords;
    TBOX box;
    blob.GetPreciseBoundingBox(&box);
    box.pad(1, 1);
    blob.GetEdgeCoords(box, &x_coords, &y_coords);
    cn_denorm->SetupNonLinear(&blob.denorm(), box, UINT8_MAX, UINT8_MAX,
                              0.0f, 0.0f, x_coords, y_coords);
  } else {
    cn_denorm->SetupNormalization(nullptr, nullptr, &blob.denorm(),
                                  center.x(), center.y(),
                                  51.2f / second_moments.x(),
                                  51.2f / second_moments.y(),
                                  128.0f, 128.0f);
  }
}

// Extract the OCR results, costs (penalty points for uncertainty),
// and the bounding boxes of the characters.
static void extract_result(ELIST_ITERATOR *out,
                           PAGE_RES* page_res) {
  PAGE_RES_IT page_res_it(page_res);
  int word_count = 0;
  while (page_res_it.word() != NULL) {
    WERD_RES *word = page_res_it.word();
    const char *str = word->best_choice->string().string();
    const char *len = word->best_choice->lengths().string();

    if (word_count)
      add_space(out);
    TBOX bln_rect;
    PBLOB_LIST *blobs = word->outword->blob_list();
    PBLOB_IT it(blobs);
    int n = strlen(len);
    TBOX** boxes_to_fix = new TBOX*[n];
    for (int i = 0; i < n; i++) {
      PBLOB *blob = it.data();
      TBOX current = pblob_get_bbox(blob);
      bln_rect.bounding_union(current);

      TESS_CHAR *tc = new TESS_CHAR(rating_to_cost(word->best_choice->rating()),
                                    str, *len);
      tc->box = current;
      boxes_to_fix[i] = &tc->box;

      out->add_after_then_move(tc);
      it.forward();
      str += *len;
      len++;
    }

    // Find the word bbox before normalization.
    // Here we can't use the C_BLOB bboxes directly,
    // since connected letters are not yet cut.
    TBOX real_rect = c_blob_list_get_bbox(word->word->cblob_list());

    // Denormalize boxes by transforming the bbox of the whole bln word
    // into the denorm bbox (`real_rect') of the whole word.
    double x_stretch = double(real_rect.width()) / bln_rect.width();
    double y_stretch = double(real_rect.height()) / bln_rect.height();
    for (int j = 0; j < n; j++) {
      TBOX *box = boxes_to_fix[j];
      int x0 = int(real_rect.left() +
                   x_stretch * (box->left() - bln_rect.left()) + 0.5);
      int x1 = int(real_rect.left() +
                   x_stretch * (box->right() - bln_rect.left()) + 0.5);
      int y0 = int(real_rect.bottom() +
                   y_stretch * (box->bottom() - bln_rect.bottom()) + 0.5);
      int y1 = int(real_rect.bottom() +
                   y_stretch * (box->top() - bln_rect.bottom()) + 0.5);
      *box = TBOX(ICOORD(x0, y0), ICOORD(x1, y1));
    }
    delete [] boxes_to_fix;

    page_res_it.forward();
    word_count++;
  }
}

/// Helper to compute the dispute resolution metric.
/// Disputed blob resolution. The aim is to give the blob to the most
/// appropriate boxfile box. Most of the time it is obvious, but if
/// two boxfile boxes overlap significantly it is not. If a small boxfile
/// box takes most of the blob, and a large boxfile box does too, then
/// we want the small boxfile box to get it, but if the small box
/// is much smaller than the blob, we don't want it to get it.
/// Details of the disputed blob resolution:
/// Given a box with area A, and a blob with area B, with overlap area C,
/// then the miss metric is (A-C)(B-C)/(AB) and the box with minimum
/// miss metric gets the blob.
static double BoxMissMetric(const TBOX& box1, const TBOX& box2) {
  int overlap_area = box1.intersection(box2).area();
  double miss_metric = box1.area()- overlap_area;
  miss_metric /= box1.area();
  miss_metric *= box2.area() - overlap_area;
  miss_metric /= box2.area();
  return miss_metric;
}