BOX* M_Utils::getColPartImCoords(ColPartition* cp, PIX* im) {
BLOBNBOX_CLIST* blobnboxes = cp->boxes();
CLIST_ITERATOR bbox_it(blobnboxes);
l_int32 height = (l_int32)im->h;
l_int32 left = INT_MAX;
l_int32 right = INT_MIN;
l_int32 top = INT_MAX;
l_int32 bottom = INT_MIN;
l_int32 l, r, t, b;
for (bbox_it.mark_cycle_pt (); !bbox_it.cycled_list();
bbox_it.forward()) {
BLOBNBOX* blobnbox = (BLOBNBOX*)bbox_it.data();
l = (l_int32)blobnbox->cblob()->bounding_box().left();
r = (l_int32)blobnbox->cblob()->bounding_box().right();
t = height - (l_int32)blobnbox->cblob()->bounding_box().top();
b = height - (l_int32)blobnbox->cblob()->bounding_box().bottom();
if(l < left)
left = l;
if(r > right)
right = r;
if(t < top)
top = t;
if(b > bottom)
bottom = b;
}
BOX* boxret = boxCreate(left, top, right-left, bottom-top);
return boxret;
}
// Helper to compute edge offsets for all the blobs on the list.
// See coutln.h for an explanation of edge offsets.
void BLOBNBOX::ComputeEdgeOffsets(Pix* thresholds, Pix* grey,
BLOBNBOX_LIST* blobs) {
int grey_height = 0;
int thr_height = 0;
int scale_factor = 1;
if (thresholds != NULL && grey != NULL) {
grey_height = pixGetHeight(grey);
thr_height = pixGetHeight(thresholds);
scale_factor =
IntCastRounded(static_cast<double>(grey_height) / thr_height);
}
BLOBNBOX_IT blob_it(blobs);
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
BLOBNBOX* blob = blob_it.data();
if (blob->cblob() != NULL) {
// Get the threshold that applies to this blob.
l_uint32 threshold = 128;
if (thresholds != NULL && grey != NULL) {
const TBOX& box = blob->cblob()->bounding_box();
// Transform the coordinates if required.
TPOINT pt((box.left() + box.right()) / 2,
(box.top() + box.bottom()) / 2);
pixGetPixel(thresholds, pt.x / scale_factor,
thr_height - 1 - pt.y / scale_factor, &threshold);
}
blob->cblob()->ComputeEdgeOffsets(threshold, grey);
}
}
}
// Inserts a list of blobs into the projection.
// Rotation is a multiple of 90 degrees to get from blob coords to
// nontext_map coords, nontext_map_box is the bounds of the nontext_map.
// Blobs are spread horizontally or vertically according to their internal
// flags, but the spreading is truncated by set pixels in the nontext_map
// and also by the horizontal rule line limits on the blobs.
void TextlineProjection::ProjectBlobs(BLOBNBOX_LIST* blobs,
const FCOORD& rotation,
const TBOX& nontext_map_box,
Pix* nontext_map) {
BLOBNBOX_IT blob_it(blobs);
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
BLOBNBOX* blob = blob_it.data();
TBOX bbox = blob->bounding_box();
ICOORD middle((bbox.left() + bbox.right()) / 2,
(bbox.bottom() + bbox.top()) / 2);
bool spreading_horizontally = PadBlobBox(blob, &bbox);
// Rotate to match the nontext_map.
bbox.rotate(rotation);
middle.rotate(rotation);
if (rotation.x() == 0.0f)
spreading_horizontally = !spreading_horizontally;
// Clip to the image before applying the increments.
bbox &= nontext_map_box; // This is in-place box intersection.
// Check for image pixels before spreading.
TruncateBoxToMissNonText(middle.x(), middle.y(), spreading_horizontally,
nontext_map, &bbox);
if (bbox.area() > 0) {
IncrementRectangle8Bit(bbox);
}
}
}
// Computes the noise_density_ by summing the number of elements in a
// neighbourhood of each grid cell.
void StrokeWidth::ComputeNoiseDensity(TO_BLOCK* block, TabFind* line_grid) {
// Run a preliminary strokewidth neighbour detection on the medium blobs.
line_grid->InsertBlobList(true, true, false, &block->blobs, false, this);
BLOBNBOX_IT blob_it(&block->blobs);
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
SetNeighbours(false, blob_it.data());
}
// Remove blobs with a good strokewidth neighbour from the grid.
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
BLOBNBOX* blob = blob_it.data();
if (blob->GoodTextBlob() > 0)
RemoveBBox(blob);
blob->ClearNeighbours();
}
// Insert the smaller blobs into the grid.
line_grid->InsertBlobList(true, true, false, &block->small_blobs,
false, this);
line_grid->InsertBlobList(true, true, false, &block->noise_blobs,
false, this);
if (noise_density_ != NULL)
delete noise_density_;
IntGrid* cell_counts = CountCellElements();
noise_density_ = cell_counts->NeighbourhoodSum();
delete cell_counts;
// Clear the grid as we don't want the small stuff hanging around in it.
Clear();
}
// Helper function to divide the input blobs over noise, small, medium
// and large lists. Blobs small in height and (small in width or large in width)
// go in the noise list. Dash (-) candidates go in the small list, and
// medium and large are by height.
// SIDE-EFFECT: reset all blobs to initial state by calling Init().
static void SizeFilterBlobs(int min_height, int max_height,
BLOBNBOX_LIST* src_list,
BLOBNBOX_LIST* noise_list,
BLOBNBOX_LIST* small_list,
BLOBNBOX_LIST* medium_list,
BLOBNBOX_LIST* large_list) {
BLOBNBOX_IT noise_it(noise_list);
BLOBNBOX_IT small_it(small_list);
BLOBNBOX_IT medium_it(medium_list);
BLOBNBOX_IT large_it(large_list);
for (BLOBNBOX_IT src_it(src_list); !src_it.empty(); src_it.forward()) {
BLOBNBOX* blob = src_it.extract();
blob->ReInit();
int width = blob->bounding_box().width();
int height = blob->bounding_box().height();
if (height < min_height &&
(width < min_height || width > max_height))
noise_it.add_after_then_move(blob);
else if (height > max_height)
large_it.add_after_then_move(blob);
else if (height < min_height)
small_it.add_after_then_move(blob);
else
medium_it.add_after_then_move(blob);
}
}
// Inserts all the blobs from the given list, with x and y spreading,
// without removing from the source list, so ownership remains with the
// source list.
void BlobGrid::InsertBlobList(BLOBNBOX_LIST * blobs) {
BLOBNBOX_IT blob_it(blobs);
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
BLOBNBOX *blob = blob_it.data();
if (!blob->joined_to_prev())
InsertBBox(true, true, blob);
}
}
static void clear_blobnboxes(BLOBNBOX_LIST* boxes) {
BLOBNBOX_IT it = boxes;
// A BLOBNBOX generally doesn't own its blobs, so if they do, you
// have to delete them explicitly.
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
BLOBNBOX* box = it.data();
delete box->cblob();
}
}
// Return the partner of this TabVector if the vector qualifies as
// being a vertical text line, otherwise NULL.
TabVector* TabVector::VerticalTextlinePartner() {
if (!partners_.singleton())
return NULL;
TabVector_C_IT partner_it(&partners_);
TabVector* partner = partner_it.data();
BLOBNBOX_C_IT box_it1(&boxes_);
BLOBNBOX_C_IT box_it2(&partner->boxes_);
// Count how many boxes are also in the other list.
// At the same time, gather the mean width and median vertical gap.
if (textord_debug_tabfind > 1) {
Print("Testing for vertical text");
partner->Print(" partner");
}
int num_matched = 0;
int num_unmatched = 0;
int total_widths = 0;
int width = startpt().x() - partner->startpt().x();
if (width < 0)
width = -width;
STATS gaps(0, width * 2);
BLOBNBOX* prev_bbox = NULL;
box_it2.mark_cycle_pt();
for (box_it1.mark_cycle_pt(); !box_it1.cycled_list(); box_it1.forward()) {
BLOBNBOX* bbox = box_it1.data();
TBOX box = bbox->bounding_box();
if (prev_bbox != NULL) {
gaps.add(box.bottom() - prev_bbox->bounding_box().top(), 1);
}
while (!box_it2.cycled_list() && box_it2.data() != bbox &&
box_it2.data()->bounding_box().bottom() < box.bottom()) {
box_it2.forward();
}
if (!box_it2.cycled_list() && box_it2.data() == bbox &&
bbox->region_type() >= BRT_UNKNOWN &&
(prev_bbox == NULL || prev_bbox->region_type() >= BRT_UNKNOWN))
++num_matched;
else
++num_unmatched;
total_widths += box.width();
prev_bbox = bbox;
}
if (num_unmatched + num_matched == 0) return NULL;
double avg_width = total_widths * 1.0 / (num_unmatched + num_matched);
double max_gap = textord_tabvector_vertical_gap_fraction * avg_width;
int min_box_match = static_cast<int>((num_matched + num_unmatched) *
textord_tabvector_vertical_box_ratio);
bool is_vertical = (gaps.get_total() > 0 &&
num_matched >= min_box_match &&
gaps.median() <= max_gap);
if (textord_debug_tabfind > 1) {
tprintf("gaps=%d, matched=%d, unmatched=%d, min_match=%d "
"median gap=%.2f, width=%.2f max_gap=%.2f Vertical=%s\n",
gaps.get_total(), num_matched, num_unmatched, min_box_match,
gaps.median(), avg_width, max_gap, is_vertical?"Yes":"No");
}
return (is_vertical) ? partner : NULL;
}
// NULLs out any neighbours that are DeletableNoise to remove references.
void BLOBNBOX::CleanNeighbours() {
for (int dir = 0; dir < BND_COUNT; ++dir) {
BLOBNBOX* neighbour = neighbours_[dir];
if (neighbour != NULL && neighbour->DeletableNoise()) {
neighbours_[dir] = NULL;
good_stroke_neighbours_[dir] = false;
}
}
}
// Helper to delete all the deletable blobs on the list.
void BLOBNBOX::DeleteNoiseBlobs(BLOBNBOX_LIST* blobs) {
BLOBNBOX_IT blob_it(blobs);
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
BLOBNBOX* blob = blob_it.data();
if (blob->DeletableNoise()) {
delete blob->cblob();
delete blob_it.extract();
}
}
}
// Print basic information about this tab vector and every box in it.
void TabVector::Debug(const char* prefix) {
Print(prefix);
BLOBNBOX_C_IT it(&boxes_);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
BLOBNBOX* bbox = it.data();
const TBOX& box = bbox->bounding_box();
tprintf("Box at (%d,%d)->(%d,%d)\n",
box.left(), box.bottom(), box.right(), box.top());
}
}
// Returns the number of side neighbours that are of type BRT_NOISE.
int BLOBNBOX::NoisyNeighbours() const {
int count = 0;
for (int dir = 0; dir < BND_COUNT; ++dir) {
BlobNeighbourDir bnd = static_cast<BlobNeighbourDir>(dir);
BLOBNBOX* blob = neighbour(bnd);
if (blob != NULL && blob->region_type() == BRT_NOISE)
++count;
}
return count;
}
// Finds horizontal line objects in the given pix.
// Uses the given resolution to determine size thresholds instead of any
// that may be present in the pix.
// The output vectors are owned by the list and Frozen (cannot refit) by
// having no boxes, as there is no need to refit or merge separator lines.
void LineFinder::FindHorizontalLines(int resolution, Pix* pix,
TabVector_LIST* vectors) {
#ifdef HAVE_LIBLEPT
Pix* line_pix;
Boxa* boxes = GetHLineBoxes(resolution, pix, &line_pix);
C_BLOB_LIST line_cblobs;
int width = pixGetWidth(pix);
int height = pixGetHeight(pix);
ConvertBoxaToBlobs(height, width, &boxes, &line_cblobs);
// Make the BLOBNBOXes from the C_BLOBs.
BLOBNBOX_LIST line_bblobs;
C_BLOB_IT blob_it(&line_cblobs);
BLOBNBOX_IT bbox_it(&line_bblobs);
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
C_BLOB* cblob = blob_it.data();
BLOBNBOX* bblob = new BLOBNBOX(cblob);
bbox_it.add_to_end(bblob);
}
ICOORD bleft(0, 0);
ICOORD tright(height, width);
int vertical_x, vertical_y;
FindLineVectors(bleft, tright, &line_bblobs, &vertical_x, &vertical_y,
vectors);
if (!vectors->empty()) {
// Some lines were found, so erase the unused blobs from the line image
// and then subtract the line image from the source.
bbox_it.move_to_first();
for (bbox_it.mark_cycle_pt(); !bbox_it.cycled_list(); bbox_it.forward()) {
BLOBNBOX* blob = bbox_it.data();
if (blob->left_tab_type() == TT_UNCONFIRMED) {
const TBOX& box = blob->bounding_box();
// Coords are in tess format so filp x and y and then covert
// to leptonica by height -y.
Box* pixbox = boxCreate(box.bottom(), height - box.right(),
box.height(), box.width());
pixClearInRect(line_pix, pixbox);
boxDestroy(&pixbox);
}
}
pixDilateBrick(line_pix, line_pix, 3, 1);
pixSubtract(pix, pix, line_pix);
if (textord_tabfind_show_vlines)
pixWrite("hlinesclean.png", line_pix, IFF_PNG);
ICOORD vertical;
vertical.set_with_shrink(vertical_x, vertical_y);
TabVector::MergeSimilarTabVectors(vertical, vectors, NULL);
// Iterate the vectors to flip them.
TabVector_IT h_it(vectors);
for (h_it.mark_cycle_pt(); !h_it.cycled_list(); h_it.forward()) {
h_it.data()->XYFlip();
}
}
pixDestroy(&line_pix);
#endif
}
// Helper to draw only DeletableNoise blobs (unowned, BRT_NOISE) on the
// given list in the given body_colour, with child outlines in the
// child_colour.
void BLOBNBOX::PlotNoiseBlobs(BLOBNBOX_LIST* list,
ScrollView::Color body_colour,
ScrollView::Color child_colour,
ScrollView* win) {
BLOBNBOX_IT it(list);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
BLOBNBOX* blob = it.data();
if (blob->DeletableNoise())
blob->plot(win, body_colour, child_colour);
}
}
// Returns the box gaps between this and its neighbours_ in an array
// indexed by BlobNeighbourDir.
void BLOBNBOX::NeighbourGaps(int gaps[BND_COUNT]) const {
for (int dir = 0; dir < BND_COUNT; ++dir) {
gaps[dir] = MAX_INT16;
BLOBNBOX* neighbour = neighbours_[dir];
if (neighbour != NULL) {
TBOX n_box = neighbour->bounding_box();
if (dir == BND_LEFT || dir == BND_RIGHT) {
gaps[dir] = box.x_gap(n_box);
} else {
gaps[dir] = box.y_gap(n_box);
}
}
}
}
// Finds vertical lines in the given list of BLOBNBOXes. bleft and tright
// are the bounds of the image on which the input line_bblobs were found.
// The input line_bblobs list is const really.
// The output vertical_x and vertical_y are the total of all the vectors.
// The output list of TabVector makes no reference to the input BLOBNBOXes.
void LineFinder::FindLineVectors(const ICOORD& bleft, const ICOORD& tright,
BLOBNBOX_LIST* line_bblobs,
int* vertical_x, int* vertical_y,
TabVector_LIST* vectors) {
BLOBNBOX_IT bbox_it(line_bblobs);
int b_count = 0;
// Put all the blobs into the grid to find the lines, and move the blobs
// to the output lists.
AlignedBlob blob_grid(kLineFindGridSize, bleft, tright);
for (bbox_it.mark_cycle_pt(); !bbox_it.cycled_list(); bbox_it.forward()) {
BLOBNBOX* bblob = bbox_it.data();
bblob->set_left_tab_type(TT_UNCONFIRMED);
bblob->set_left_rule(bleft.x());
bblob->set_right_rule(tright.x());
bblob->set_left_crossing_rule(bleft.x());
bblob->set_right_crossing_rule(tright.x());
blob_grid.InsertBBox(false, true, bblob);
++b_count;
}
if (textord_debug_tabfind)
tprintf("Inserted %d line blobs into grid\n", b_count);
if (b_count == 0)
return;
// Search the entire grid, looking for vertical line vectors.
GridSearch<BLOBNBOX, BLOBNBOX_CLIST, BLOBNBOX_C_IT> lsearch(&blob_grid);
BLOBNBOX* bbox;
TabVector_IT vector_it(vectors);
*vertical_x = 0;
*vertical_y = 1;
lsearch.StartFullSearch();
while ((bbox = lsearch.NextFullSearch()) != NULL) {
if (bbox->left_tab_type() == TT_UNCONFIRMED) {
const TBOX& box = bbox->bounding_box();
if (AlignedBlob::WithinTestRegion(2, box.left(), box.bottom()))
tprintf("Finding line vector starting at bbox (%d,%d)\n",
box.left(), box.bottom());
AlignedBlobParams align_params(*vertical_x, *vertical_y, box.width());
TabVector* vector = blob_grid.FindVerticalAlignment(align_params, bbox,
vertical_x,
vertical_y);
if (vector != NULL) {
vector->Freeze();
vector_it.add_to_end(vector);
}
}
}
ScrollView* line_win = NULL;
if (textord_tabfind_show_vlines) {
line_win = blob_grid.MakeWindow(0, 50, "Vlines");
blob_grid.DisplayBoxes(line_win);
line_win = blob_grid.DisplayTabs("Vlines", line_win);
}
}
// Moves blobs that look like they don't sit well on a textline from the
// input blobs list to the output small_blobs list.
// This gets them away from initial textline finding to stop diacritics
// from forming incorrect textlines. (Introduced mainly to fix Thai.)
void TextlineProjection::MoveNonTextlineBlobs(
BLOBNBOX_LIST* blobs, BLOBNBOX_LIST* small_blobs) const {
BLOBNBOX_IT it(blobs);
BLOBNBOX_IT small_it(small_blobs);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
BLOBNBOX* blob = it.data();
const TBOX& box = blob->bounding_box();
bool debug = AlignedBlob::WithinTestRegion(2, box.left(),
box.bottom());
if (BoxOutOfHTextline(box, NULL, debug) && !blob->UniquelyVertical()) {
blob->ClearNeighbours();
small_it.add_to_end(it.extract());
}
}
}
// Returns true if other has a similar stroke width to this.
bool BLOBNBOX::MatchingStrokeWidth(const BLOBNBOX& other,
double fractional_tolerance,
double constant_tolerance) const {
// The perimeter-based width is used as a backup in case there is
// no information in the blob.
double p_width = area_stroke_width();
double n_p_width = other.area_stroke_width();
float h_tolerance = horz_stroke_width_ * fractional_tolerance
+ constant_tolerance;
float v_tolerance = vert_stroke_width_ * fractional_tolerance
+ constant_tolerance;
double p_tolerance = p_width * fractional_tolerance
+ constant_tolerance;
bool h_zero = horz_stroke_width_ == 0.0f || other.horz_stroke_width_ == 0.0f;
bool v_zero = vert_stroke_width_ == 0.0f || other.vert_stroke_width_ == 0.0f;
bool h_ok = !h_zero && NearlyEqual(horz_stroke_width_,
other.horz_stroke_width_, h_tolerance);
bool v_ok = !v_zero && NearlyEqual(vert_stroke_width_,
other.vert_stroke_width_, v_tolerance);
bool p_ok = h_zero && v_zero && NearlyEqual(p_width, n_p_width, p_tolerance);
// For a match, at least one of the horizontal and vertical widths
// must match, and the other one must either match or be zero.
// Only if both are zero will we look at the perimeter metric.
return p_ok || ((v_ok || h_ok) && (h_ok || h_zero) && (v_ok || v_zero));
}
TBOX box_next_pre_chopped( //get bounding box
BLOBNBOX_IT *it //iterator to blobds
) {
BLOBNBOX *blob; //current blob
TBOX result; //total box
blob = it->data ();
result = blob->bounding_box ();
do {
it->forward ();
blob = it->data ();
}
//until next real blob
while (blob->joined_to_prev ());
return result;
}
// Finds vertical line objects in the given pix.
// Uses the given resolution to determine size thresholds instead of any
// that may be present in the pix.
// The output vertical_x and vertical_y contain a sum of the output vectors,
// thereby giving the mean vertical direction.
// The output vectors are owned by the list and Frozen (cannot refit) by
// having no boxes, as there is no need to refit or merge separator lines.
void LineFinder::FindVerticalLines(int resolution, Pix* pix,
int* vertical_x, int* vertical_y,
TabVector_LIST* vectors) {
#ifdef HAVE_LIBLEPT
Pix* line_pix;
Boxa* boxes = GetVLineBoxes(resolution, pix, &line_pix);
C_BLOB_LIST line_cblobs;
int width = pixGetWidth(pix);
int height = pixGetHeight(pix);
ConvertBoxaToBlobs(width, height, &boxes, &line_cblobs);
// Make the BLOBNBOXes from the C_BLOBs.
BLOBNBOX_LIST line_bblobs;
C_BLOB_IT blob_it(&line_cblobs);
BLOBNBOX_IT bbox_it(&line_bblobs);
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
C_BLOB* cblob = blob_it.data();
BLOBNBOX* bblob = new BLOBNBOX(cblob);
bbox_it.add_to_end(bblob);
}
ICOORD bleft(0, 0);
ICOORD tright(width, height);
FindLineVectors(bleft, tright, &line_bblobs, vertical_x, vertical_y, vectors);
if (!vectors->empty()) {
// Some lines were found, so erase the unused blobs from the line image
// and then subtract the line image from the source.
bbox_it.move_to_first();
for (bbox_it.mark_cycle_pt(); !bbox_it.cycled_list(); bbox_it.forward()) {
BLOBNBOX* blob = bbox_it.data();
if (blob->left_tab_type() == TT_UNCONFIRMED) {
const TBOX& box = blob->bounding_box();
Box* pixbox = boxCreate(box.left(), height - box.top(),
box.width(), box.height());
pixClearInRect(line_pix, pixbox);
boxDestroy(&pixbox);
}
}
pixDilateBrick(line_pix, line_pix, 1, 3);
pixSubtract(pix, pix, line_pix);
if (textord_tabfind_show_vlines)
pixWrite("vlinesclean.png", line_pix, IFF_PNG);
ICOORD vertical;
vertical.set_with_shrink(*vertical_x, *vertical_y);
TabVector::MergeSimilarTabVectors(vertical, vectors, NULL);
}
pixDestroy(&line_pix);
#endif
}
// Display the blobs in the window colored according to textline quality.
void TextlineProjection::PlotGradedBlobs(BLOBNBOX_LIST* blobs,
ScrollView* win) {
#ifndef GRAPHICS_DISABLED
BLOBNBOX_IT it(blobs);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
BLOBNBOX* blob = it.data();
const TBOX& box = blob->bounding_box();
bool bad_box = BoxOutOfHTextline(box, NULL, false);
if (blob->UniquelyVertical())
win->Pen(ScrollView::YELLOW);
else
win->Pen(bad_box ? ScrollView::RED : ScrollView::BLUE);
win->Rectangle(box.left(), box.bottom(), box.right(), box.top());
}
win->Update();
#endif // GRAPHICS_DISABLED
}
// Display the tab codes of the BLOBNBOXes in this grid.
ScrollView* AlignedBlob::DisplayTabs(const char* window_name,
ScrollView* tab_win) {
#ifndef GRAPHICS_DISABLED
if (tab_win == nullptr)
tab_win = MakeWindow(0, 50, window_name);
// For every tab in the grid, display it.
GridSearch<BLOBNBOX, BLOBNBOX_CLIST, BLOBNBOX_C_IT> gsearch(this);
gsearch.StartFullSearch();
BLOBNBOX* bbox;
while ((bbox = gsearch.NextFullSearch()) != nullptr) {
const TBOX& box = bbox->bounding_box();
int left_x = box.left();
int right_x = box.right();
int top_y = box.top();
int bottom_y = box.bottom();
TabType tabtype = bbox->left_tab_type();
if (tabtype != TT_NONE) {
if (tabtype == TT_MAYBE_ALIGNED)
tab_win->Pen(ScrollView::BLUE);
else if (tabtype == TT_MAYBE_RAGGED)
tab_win->Pen(ScrollView::YELLOW);
else if (tabtype == TT_CONFIRMED)
tab_win->Pen(ScrollView::GREEN);
else
tab_win->Pen(ScrollView::GREY);
tab_win->Line(left_x, top_y, left_x, bottom_y);
}
tabtype = bbox->right_tab_type();
if (tabtype != TT_NONE) {
if (tabtype == TT_MAYBE_ALIGNED)
tab_win->Pen(ScrollView::MAGENTA);
else if (tabtype == TT_MAYBE_RAGGED)
tab_win->Pen(ScrollView::ORANGE);
else if (tabtype == TT_CONFIRMED)
tab_win->Pen(ScrollView::RED);
else
tab_win->Pen(ScrollView::GREY);
tab_win->Line(right_x, top_y, right_x, bottom_y);
}
}
tab_win->Update();
#endif
return tab_win;
}
// Sets up displacement_modes_ with the top few modes of the perpendicular
// distance of each blob from the given direction vector, after rounding.
void BaselineRow::SetupBlobDisplacements(const FCOORD& direction) {
// Set of perpendicular displacements of the blob bottoms from the required
// baseline direction.
GenericVector<double> perp_blob_dists;
displacement_modes_.truncate(0);
// Gather the skew-corrected position of every blob.
double min_dist = MAX_FLOAT32;
double max_dist = -MAX_FLOAT32;
BLOBNBOX_IT blob_it(blobs_);
bool debug = false;
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
BLOBNBOX* blob = blob_it.data();
const TBOX& box = blob->bounding_box();
#ifdef kDebugYCoord
if (box.bottom() < kDebugYCoord && box.top() > kDebugYCoord) debug = true;
#endif
FCOORD blob_pos((box.left() + box.right()) / 2.0f,
blob->baseline_position());
double offset = direction * blob_pos;
perp_blob_dists.push_back(offset);
if (debug) {
tprintf("Displacement %g for blob at:", offset);
box.print();
}
UpdateRange(offset, &min_dist, &max_dist);
}
// Set up a histogram using disp_quant_factor_ as the bucket size.
STATS dist_stats(IntCastRounded(min_dist / disp_quant_factor_),
IntCastRounded(max_dist / disp_quant_factor_) + 1);
for (int i = 0; i < perp_blob_dists.size(); ++i) {
dist_stats.add(IntCastRounded(perp_blob_dists[i] / disp_quant_factor_), 1);
}
GenericVector<KDPairInc<float, int> > scaled_modes;
dist_stats.top_n_modes(kMaxDisplacementsModes, &scaled_modes);
if (debug) {
for (int i = 0; i < scaled_modes.size(); ++i) {
tprintf("Top mode = %g * %d\n",
scaled_modes[i].key * disp_quant_factor_, scaled_modes[i].data);
}
}
for (int i = 0; i < scaled_modes.size(); ++i)
displacement_modes_.push_back(disp_quant_factor_ * scaled_modes[i].key);
}
// Returns true if the given blob overlaps more than max_overlaps blobs
// in the current grid.
bool CCNonTextDetect::BlobOverlapsTooMuch(BLOBNBOX* blob, int max_overlaps) {
// Search the grid to see what intersects it.
// Setup a Rectangle search for overlapping this blob.
BlobGridSearch rsearch(this);
TBOX box = blob->bounding_box();
rsearch.StartRectSearch(box);
rsearch.SetUniqueMode(true);
BLOBNBOX* neighbour;
int overlap_count = 0;
while (overlap_count <= max_overlaps &&
(neighbour = rsearch.NextRectSearch()) != NULL) {
if (box.major_overlap(neighbour->bounding_box())) {
++overlap_count;
if (overlap_count > max_overlaps)
return true;
}
}
return false;
}
//yangjing01 modified :
bool TAL_make_single_word(bool one_blob, TO_ROW_LIST* rows, ROW_LIST* real_rows)
{
TO_ROW_IT to_row_it(rows);
ROW_IT row_it(real_rows);
//to_real_row is the real row information of single row or single char mode
TO_ROW* real_to_row = NULL;
float row_max_height = 0.0;
for (to_row_it.mark_cycle_pt();
!to_row_it.cycled_list(); to_row_it.forward()){
TO_ROW* row = to_row_it.data();
float row_min_y = row->min_y();
float row_max_y = row->max_y();
float row_height = abs(row_max_y - row_min_y);
if (real_to_row == NULL
|| row_height > row_max_height
|| fabs(row_height - row_max_height) < 1.0f){
row_max_height = row_height;
real_to_row = row;
}
}
if (real_to_row == NULL){
return false;
}
C_BLOB_LIST cblobs;
C_BLOB_IT cblob_it(&cblobs);
BLOBNBOX_IT box_it(real_to_row->blob_list());
for (; !box_it.empty(); box_it.forward()){
BLOBNBOX* bblob = box_it.extract();
if (bblob->joined_to_prev() || (one_blob && !cblob_it.empty())) {
if (bblob->cblob() != NULL){
C_OUTLINE_IT cout_it(cblob_it.data()->out_list());
cout_it.move_to_last();
cout_it.add_list_after(bblob->cblob()->out_list());
delete bblob->cblob();
}
}
else {
if (bblob->cblob() != NULL)
cblob_it.add_after_then_move(bblob->cblob());
}
delete bblob;
}
// Convert the TO_ROW to a ROW.
ROW* real_row = new ROW(real_to_row, static_cast<inT16>(real_to_row->kern_size),
static_cast<inT16>(real_to_row->space_size));
WERD_IT word_it(real_row->word_list());
WERD* word = new WERD(&cblobs, 0, NULL);
word->set_flag(W_BOL, TRUE);
word->set_flag(W_EOL, TRUE);
word->set_flag(W_DONT_CHOP, one_blob);
word_it.add_after_then_move(word);
row_it.add_after_then_move(real_row);
return true;
}
// Extend this vector to include the supplied blob if it doesn't
// already have it.
void TabVector::ExtendToBox(BLOBNBOX* new_blob) {
TBOX new_box = new_blob->bounding_box();
BLOBNBOX_C_IT it(&boxes_);
if (!it.empty()) {
BLOBNBOX* blob = it.data();
TBOX box = blob->bounding_box();
while (!it.at_last() && box.top() <= new_box.top()) {
if (blob == new_blob)
return; // We have it already.
it.forward();
blob = it.data();
box = blob->bounding_box();
}
if (box.top() >= new_box.top()) {
it.add_before_stay_put(new_blob);
needs_refit_ = true;
return;
}
}
needs_refit_ = true;
it.add_after_stay_put(new_blob);
}
void TO_ROW::compute_vertical_projection() { //project whole row
TBOX row_box; //bound of row
BLOBNBOX *blob; //current blob
TBOX blob_box; //bounding box
BLOBNBOX_IT blob_it = blob_list ();
if (blob_it.empty ())
return;
row_box = blob_it.data ()->bounding_box ();
for (blob_it.mark_cycle_pt (); !blob_it.cycled_list (); blob_it.forward ())
row_box += blob_it.data ()->bounding_box ();
projection.set_range (row_box.left () - PROJECTION_MARGIN,
row_box.right () + PROJECTION_MARGIN);
projection_left = row_box.left () - PROJECTION_MARGIN;
projection_right = row_box.right () + PROJECTION_MARGIN;
for (blob_it.mark_cycle_pt (); !blob_it.cycled_list (); blob_it.forward ()) {
blob = blob_it.data();
if (blob->cblob() != NULL)
vertical_cblob_projection(blob->cblob(), &projection);
}
}
// 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());
}
}
}
BOX box_next( //get bounding box
BLOBNBOX_IT *it //iterator to blobds
) {
BLOBNBOX *blob; //current blob
BOX result; //total box
blob = it->data ();
result = blob->bounding_box ();
do {
it->forward ();
blob = it->data ();
if (blob->blob () == NULL && blob->cblob () == NULL)
//was pre-chopped
result += blob->bounding_box ();
}
//until next real blob
while (blob->blob () == NULL && blob->cblob () == NULL || blob->joined_to_prev ());
return result;
}
WERD *make_real_word(BLOBNBOX_IT *box_it, //iterator
int32_t blobcount, //no of blobs to use
bool bol, //start of line
uint8_t blanks //no of blanks
) {
C_OUTLINE_IT cout_it;
C_BLOB_LIST cblobs;
C_BLOB_IT cblob_it = &cblobs;
WERD *word; // new word
BLOBNBOX *bblob; // current blob
int32_t blobindex; // in row
for (blobindex = 0; blobindex < blobcount; blobindex++) {
bblob = box_it->extract();
if (bblob->joined_to_prev()) {
if (bblob->cblob() != nullptr) {
cout_it.set_to_list(cblob_it.data()->out_list());
cout_it.move_to_last();
cout_it.add_list_after(bblob->cblob()->out_list());
delete bblob->cblob();
}
}
else {
if (bblob->cblob() != nullptr)
cblob_it.add_after_then_move(bblob->cblob());
}
delete bblob;
box_it->forward(); // next one
}
if (blanks < 1)
blanks = 1;
word = new WERD(&cblobs, blanks, nullptr);
if (bol)
word->set_flag(W_BOL, true);
if (box_it->at_first())
word->set_flag(W_EOL, true); // at end of line
return word;
}
