// Copies the given feature_space and uses it as the index feature map
// from INT_FEATURE_STRUCT.
void IntFeatureMap::Init(const IntFeatureSpace& feature_space) {
feature_space_ = feature_space;
mapping_changed_ = false;
int sparse_size = feature_space_.Size();
feature_map_.Init(sparse_size, true);
feature_map_.Setup();
compact_size_ = feature_map_.CompactSize();
// Initialize look-up tables if needed.
FCOORD dir = FeatureDirection(0);
if (dir.x() == 0.0f && dir.y() == 0.0f)
InitIntegerFX();
// Compute look-up tables to generate offset features.
for (int dir = 0; dir < kNumOffsetMaps; ++dir) {
delete [] offset_plus_[dir];
delete [] offset_minus_[dir];
offset_plus_[dir] = new int[sparse_size];
offset_minus_[dir] = new int[sparse_size];
}
for (int dir = 1; dir <= kNumOffsetMaps; ++dir) {
for (int i = 0; i < sparse_size; ++i) {
int offset_index = ComputeOffsetFeature(i, dir);
offset_plus_[dir - 1][i] = offset_index;
offset_index = ComputeOffsetFeature(i, -dir);
offset_minus_[dir - 1][i] = offset_index;
}
}
}
// Rotate the grid by rotation, keeping cell contents.
// rotation must be a multiple of 90 degrees.
// NOTE: due to partial cells, cell coverage in the rotated grid will be
// inexact. This is why there is no Rotate for the generic BBGrid.
// TODO(rays) investigate fixing this inaccuracy by moving the origin after
// rotation.
void IntGrid::Rotate(const FCOORD& rotation) {
ASSERT_HOST(rotation.x() == 0.0f || rotation.y() == 0.0f);
ICOORD old_bleft(bleft());
ICOORD old_tright(tright());
int old_width = gridwidth();
int old_height = gridheight();
TBOX box(bleft(), tright());
box.rotate(rotation);
int* old_grid = grid_;
grid_ = NULL;
Init(gridsize(), box.botleft(), box.topright());
// Iterate over the old grid, copying data to the rotated position in the new.
int oldi = 0;
FCOORD x_step(rotation);
x_step *= gridsize();
for (int oldy = 0; oldy < old_height; ++oldy) {
FCOORD line_pos(old_bleft.x(), old_bleft.y() + gridsize() * oldy);
line_pos.rotate(rotation);
for (int oldx = 0; oldx < old_width; ++oldx, line_pos += x_step, ++oldi) {
int grid_x, grid_y;
GridCoords(static_cast<int>(line_pos.x() + 0.5),
static_cast<int>(line_pos.y() + 0.5),
&grid_x, &grid_y);
grid_[grid_y * gridwidth() + grid_x] = old_grid[oldi];
}
}
delete [] old_grid;
}
inT16 OUTLINE::winding_number( //winding number
const FCOORD &point //point to wind around
) {
inT16 count; //winding count
POLYPT *polypt; //current point
FCOORD vec; //to current point
float cross; //cross product
POLYPT_IT it = &outline; //iterator
count = 0;
do {
polypt = it.data ();
vec = polypt->pos - point;
//crossing the line
if (vec.y () <= 0 && vec.y () + polypt->vec.y () > 0) {
cross = vec * polypt->vec; //cross product
if (cross > 0)
count++; //crossing right half
else if (cross == 0)
return INTERSECTING; //going through point
}
else if (vec.y () > 0 && vec.y () + polypt->vec.y () <= 0) {
cross = vec * polypt->vec;
if (cross < 0)
count--; //crossing back
else if (cross == 0)
return INTERSECTING; //illegal
}
it.forward ();
}
while (!it.at_first ());
return count; //winding number
}
void DENORM::LocalNormTransform(const FCOORD& pt, FCOORD* transformed) const {
FCOORD translated(pt.x() - x_origin_, pt.y() - YOriginAtOrigX(pt.x()));
translated.set_x(translated.x() * x_scale_);
translated.set_y(translated.y() * YScaleAtOrigX(pt.x()));
if (rotation_ != NULL)
translated.rotate(*rotation_);
transformed->set_x(translated.x() + final_xshift_);
transformed->set_y(translated.y() + final_yshift_);
}
// Rotates by the given rotation in place.
void TESSLINE::Rotate(const FCOORD rot) {
EDGEPT* pt = loop;
do {
int tmp = static_cast<int>(floor(pt->pos.x * rot.x() -
pt->pos.y * rot.y() + 0.5));
pt->pos.y = static_cast<int>(floor(pt->pos.y * rot.x() +
pt->pos.x * rot.y() + 0.5));
pt->pos.x = tmp;
pt = pt->next;
} while (pt != loop);
SetupFromPos();
}
void DENORM::LocalDenormTransform(const FCOORD& pt, FCOORD* original) const {
FCOORD rotated(pt.x() - final_xshift_, pt.y() - final_yshift_);
if (rotation_ != NULL) {
FCOORD inverse_rotation(rotation_->x(), -rotation_->y());
rotated.rotate(inverse_rotation);
}
original->set_x(rotated.x() / x_scale_ + x_origin_);
float y_scale = y_scale_;
if (num_segs_ > 0)
y_scale = YScaleAtOrigX(original->x());
original->set_y(rotated.y() / y_scale + YOriginAtOrigX(original->x()));
}
// Gathers outline points and their directions from start_index into dirs by
// stepping along the outline and normalizing the coordinates until the
// required feature_length has been collected or end_index is reached.
// On input pos must point to the position corresponding to start_index and on
// return pos is updated to the current raw position, and pos_normed is set to
// the normed version of pos.
// Since directions wrap-around, they need special treatment to get the mean.
// Provided the cluster of directions doesn't straddle the wrap-around point,
// the simple mean works. If they do, then, unless the directions are wildly
// varying, the cluster rotated by 180 degrees will not straddle the wrap-
// around point, so mean(dir + 180 degrees) - 180 degrees will work. Since
// LLSQ conveniently stores the mean of 2 variables, we use it to store
// dir and dir+128 (128 is 180 degrees) and then use the resulting mean
// with the least variance.
static int GatherPoints(const C_OUTLINE* outline, double feature_length,
const DENORM& denorm, const DENORM* root_denorm,
int start_index, int end_index,
ICOORD* pos, FCOORD* pos_normed,
LLSQ* points, LLSQ* dirs) {
int step_length = outline->pathlength();
ICOORD step = outline->step(start_index % step_length);
// Prev_normed is the start point of this collection and will be set on the
// first iteration, and on later iterations used to determine the length
// that has been collected.
FCOORD prev_normed;
points->clear();
dirs->clear();
int num_points = 0;
int index;
for (index = start_index; index <= end_index; ++index, *pos += step) {
step = outline->step(index % step_length);
int edge_weight = outline->edge_strength_at_index(index % step_length);
if (edge_weight == 0) {
// This point has conflicting gradient and step direction, so ignore it.
continue;
}
// Get the sub-pixel precise location and normalize.
FCOORD f_pos = outline->sub_pixel_pos_at_index(*pos, index % step_length);
denorm.NormTransform(root_denorm, f_pos, pos_normed);
if (num_points == 0) {
// The start of this segment.
prev_normed = *pos_normed;
} else {
FCOORD offset = *pos_normed - prev_normed;
float length = offset.length();
if (length > feature_length) {
// We have gone far enough from the start. We will use this point in
// the next set so return what we have so far.
return index;
}
}
points->add(pos_normed->x(), pos_normed->y(), edge_weight);
int direction = outline->direction_at_index(index % step_length);
if (direction >= 0) {
direction = NormalizeDirection(direction, f_pos, denorm, root_denorm);
// Use both the direction and direction +128 so we are not trying to
// take the mean of something straddling the wrap-around point.
dirs->add(direction, Modulo(direction + 128, 256));
}
++num_points;
}
return index;
}
void DENORM::LocalNormTransform(const FCOORD& pt, FCOORD* transformed) const {
FCOORD translated(pt.x() - x_origin_, pt.y() - y_origin_);
if (x_map_ != NULL && y_map_ != NULL) {
int x = ClipToRange(IntCastRounded(translated.x()), 0, x_map_->size()-1);
translated.set_x((*x_map_)[x]);
int y = ClipToRange(IntCastRounded(translated.y()), 0, y_map_->size()-1);
translated.set_y((*y_map_)[y]);
} else {
translated.set_x(translated.x() * x_scale_);
translated.set_y(translated.y() * y_scale_);
if (rotation_ != NULL)
translated.rotate(*rotation_);
}
transformed->set_x(translated.x() + final_xshift_);
transformed->set_y(translated.y() + final_yshift_);
}
// 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);
}
}
}
void DENORM::LocalDenormTransform(const FCOORD& pt, FCOORD* original) const {
FCOORD rotated(pt.x() - final_xshift_, pt.y() - final_yshift_);
if (x_map_ != NULL && y_map_ != NULL) {
int x = x_map_->binary_search(rotated.x());
original->set_x(x + x_origin_);
int y = y_map_->binary_search(rotated.y());
original->set_y(y + y_origin_);
} else {
if (rotation_ != NULL) {
FCOORD inverse_rotation(rotation_->x(), -rotation_->y());
rotated.rotate(inverse_rotation);
}
original->set_x(rotated.x() / x_scale_ + x_origin_);
float y_scale = y_scale_;
original->set_y(rotated.y() / y_scale + y_origin_);
}
}
void PageIterator::Orientation(tesseract::Orientation *orientation,
tesseract::WritingDirection *writing_direction,
tesseract::TextlineOrder *textline_order,
float *deskew_angle) const {
BLOCK* block = it_->block()->block;
// Orientation
FCOORD up_in_image(0.0, 1.0);
up_in_image.unrotate(block->classify_rotation());
up_in_image.rotate(block->re_rotation());
if (up_in_image.x() == 0.0F) {
if (up_in_image.y() > 0.0F) {
*orientation = ORIENTATION_PAGE_UP;
} else {
*orientation = ORIENTATION_PAGE_DOWN;
}
} else if (up_in_image.x() > 0.0F) {
*orientation = ORIENTATION_PAGE_RIGHT;
} else {
*orientation = ORIENTATION_PAGE_LEFT;
}
// Writing direction
bool is_vertical_text = (block->classify_rotation().x() == 0.0);
bool right_to_left = block->right_to_left();
*writing_direction =
is_vertical_text
? WRITING_DIRECTION_TOP_TO_BOTTOM
: (right_to_left
? WRITING_DIRECTION_RIGHT_TO_LEFT
: WRITING_DIRECTION_LEFT_TO_RIGHT);
// Textline Order
bool is_mongolian = false; // TODO(eger): fix me
*textline_order = is_vertical_text
? (is_mongolian
? TEXTLINE_ORDER_LEFT_TO_RIGHT
: TEXTLINE_ORDER_RIGHT_TO_LEFT)
: TEXTLINE_ORDER_TOP_TO_BOTTOM;
// Deskew angle
FCOORD skew = block->skew(); // true horizontal for textlines
*deskew_angle = -skew.angle();
}
// Fits a line in the given direction to blobs that are close to the given
// target_offset perpendicular displacement from the direction. The fit
// error is allowed to be cheat_allowance worse than the existing fit, and
// will still be used.
// If cheat_allowance > 0, the new fit will be good and replace the current
// fit if it has better fit (with cheat) OR its error is below
// max_baseline_error_ and the old fit is marked bad.
// Otherwise the new fit will only replace the old if it is really better,
// or the old fit is marked bad and the new fit has sufficient points, as
// well as being within the max_baseline_error_.
void BaselineRow::FitConstrainedIfBetter(int debug,
const FCOORD& direction,
double cheat_allowance,
double target_offset) {
double halfrange = fit_halfrange_ * direction.length();
double min_dist = target_offset - halfrange;
double max_dist = target_offset + halfrange;
ICOORD line_pt;
double new_error = fitter_.ConstrainedFit(direction, min_dist, max_dist,
debug > 2, &line_pt);
// Allow cheat_allowance off the new error
new_error -= cheat_allowance;
double old_angle = BaselineAngle();
double new_angle = direction.angle();
if (debug > 1) {
tprintf("Constrained error = %g, original = %g",
new_error, baseline_error_);
tprintf(" angles = %g, %g, delta=%g vs threshold %g\n",
old_angle, new_angle,
new_angle - old_angle, kMaxSkewDeviation);
}
bool new_good_baseline = new_error <= max_baseline_error_ &&
(cheat_allowance > 0.0 || fitter_.SufficientPointsForIndependentFit());
// The new will replace the old if any are true:
// 1. the new error is better
// 2. the old is NOT good, but the new is
// 3. there is a wild angular difference between them (assuming that the new
// is a better guess at the angle.)
if (new_error <= baseline_error_ ||
(!good_baseline_ && new_good_baseline) ||
fabs(new_angle - old_angle) > kMaxSkewDeviation) {
baseline_error_ = new_error;
baseline_pt1_ = line_pt;
baseline_pt2_ = baseline_pt1_ + direction;
good_baseline_ = new_good_baseline;
if (debug > 1) {
tprintf("Replacing with constrained baseline, good = %d\n",
good_baseline_);
}
} else if (debug > 1) {
tprintf("Keeping old baseline\n");
}
}
// Returns the point on the given line nearest to this, ie the point such
// that the vector point->this is perpendicular to the line.
// The line is defined as a line_point and a dir_vector for its direction.
FCOORD FCOORD::nearest_pt_on_line(const FCOORD& line_point,
const FCOORD& dir_vector) const {
FCOORD point_vector(*this - line_point);
// The dot product (%) is |dir_vector||point_vector|cos theta, so dividing by
// the square of the length of dir_vector gives us the fraction of dir_vector
// to add to line1 to get the appropriate point, so
// result = line1 + lambda dir_vector.
double lambda = point_vector % dir_vector / dir_vector.sqlength();
return line_point + (dir_vector * lambda);
}
void OUTLINE::move( // reposition OUTLINE
const FCOORD vec // by vector
) {
//child outline itertr
OUTLINE_IT child_it(&children);
POLYPT_IT poly_it(&outline); //outline point itertr
box.move (vec);
start.set_x ((inT16) floor (start.x () + vec.x () + 0.5));
// ?? Why ICOORD?
start.set_y ((inT16) floor (start.y () + vec.y () + 0.5));
// ?? Why ICOORD?
for (poly_it.mark_cycle_pt (); !poly_it.cycled_list (); poly_it.forward ())
poly_it.data ()->pos += vec;
for (child_it.mark_cycle_pt (); !child_it.cycled_list ();
child_it.forward ())
child_it.data ()->move (vec); // move child outlines
}
void OUTLINE::scale( // scale OUTLINE
const FCOORD vector //by fcoord
) {
//child outline itertr
OUTLINE_IT child_it(&children);
POLYPT_IT poly_it(&outline); //outline point itertr
POLYPT *pt;
box.scale (vector);
start.set_x ((inT16) floor (start.x () * vector.x () + 0.5));
// ?? Why ICOORD?
start.set_y ((inT16) floor (start.y () * vector.y () + 0.5));
// ?? Why ICOORD?
for (poly_it.mark_cycle_pt (); !poly_it.cycled_list (); poly_it.forward ()) {
pt = poly_it.data ();
pt->pos =
FCOORD (pt->pos.x () * vector.x (), pt->pos.y () * vector.y ());
pt->vec =
FCOORD (pt->vec.x () * vector.x (), pt->vec.y () * vector.y ());
}
for (child_it.mark_cycle_pt (); !child_it.cycled_list ();
child_it.forward ())
//scale child outlines
child_it.data ()->scale (vector);
}
DIR128::DIR128( //from fcoord
const FCOORD fc //vector to quantize
) {
int high, low, current; //binary search
low = 0;
if (fc.y () == 0) {
if (fc.x () >= 0)
dir = 0;
else
dir = MODULUS / 2;
return;
}
high = MODULUS;
do {
current = (high + low) / 2;
if (dirtab[current] * fc >= 0)
low = current;
else
high = current;
}
while (high - low > 1);
dir = low;
}
// Rotates the box by the angle given by rotation.
// If the blob is a diacritic, then only small rotations for skew
// correction can be applied.
void BLOBNBOX::rotate_box(FCOORD rotation) {
if (IsDiacritic()) {
ASSERT_HOST(rotation.x() >= kCosSmallAngle)
ICOORD top_pt((box.left() + box.right()) / 2, base_char_top_);
ICOORD bottom_pt(top_pt.x(), base_char_bottom_);
top_pt.rotate(rotation);
base_char_top_ = top_pt.y();
bottom_pt.rotate(rotation);
base_char_bottom_ = bottom_pt.y();
box.rotate(rotation);
} else {
box.rotate(rotation);
set_diacritic_box(box);
}
}
// Draws the features in the given window.
void WordFeature::Draw(const GenericVector<WordFeature>& features,
ScrollView* window) {
for (int f = 0; f < features.size(); ++f) {
FCOORD pos(features[f].x_, features[f].y_);
FCOORD dir;
dir.from_direction(features[f].dir_);
dir *= 8.0f;
window->SetCursor(IntCastRounded(pos.x() - dir.x()),
IntCastRounded(pos.y() - dir.y()));
window->DrawTo(IntCastRounded(pos.x() + dir.x()),
IntCastRounded(pos.y() + dir.y()));
}
}
void POLY_BLOCK::rotate(FCOORD rotation) {
FCOORD pos; //current pos;
ICOORDELT *pt; //current point
ICOORDELT_IT pts = &vertices; //iterator
do {
pt = pts.data ();
pos.set_x (pt->x ());
pos.set_y (pt->y ());
pos.rotate (rotation);
pt->set_x ((inT16) (floor (pos.x () + 0.5)));
pt->set_y ((inT16) (floor (pos.y () + 0.5)));
pts.forward ();
}
while (!pts.at_first ());
compute_bb();
}
// Adds any edges from a single segment of outline between pt1 and pt2 to
// the x_coords, y_coords vectors. pt1 and pt2 should be relative to the
// bottom-left of the bounding box, hence indices to x_coords, y_coords
// are clipped to ([0,x_limit], [0,y_limit]).
// See GetEdgeCoords above for a description of x_coords, y_coords.
static void SegmentCoords(const FCOORD& pt1, const FCOORD& pt2,
int x_limit, int y_limit,
GenericVector<GenericVector<int> >* x_coords,
GenericVector<GenericVector<int> >* y_coords) {
FCOORD step(pt2);
step -= pt1;
int start = ClipToRange(IntCastRounded(MIN(pt1.x(), pt2.x())), 0, x_limit);
int end = ClipToRange(IntCastRounded(MAX(pt1.x(), pt2.x())), 0, x_limit);
for (int x = start; x < end; ++x) {
int y = IntCastRounded(pt1.y() + step.y() * (x + 0.5 - pt1.x()) / step.x());
(*y_coords)[x].push_back(y);
}
start = ClipToRange(IntCastRounded(MIN(pt1.y(), pt2.y())), 0, y_limit);
end = ClipToRange(IntCastRounded(MAX(pt1.y(), pt2.y())), 0, y_limit);
for (int y = start; y < end; ++y) {
int x = IntCastRounded(pt1.x() + step.x() * (y + 0.5 - pt1.y()) / step.y());
(*x_coords)[y].push_back(x);
}
}
/*************************************************************************
* PIXROW::PIXROW()
*
* Constructor for a specified size PIXROW from a blob
*************************************************************************/
PIXROW::PIXROW(INT16 pos, INT16 count, PBLOB *blob) {
OUTLINE_LIST *outline_list;
OUTLINE_IT outline_it;
POLYPT_LIST *pts_list;
POLYPT_IT pts_it;
INT16 i;
FCOORD pt;
FCOORD vec;
float y_coord;
INT16 x_coord;
row_offset = pos;
row_count = count;
min = (INT16 *) alloc_mem (count * sizeof (INT16));
max = (INT16 *) alloc_mem (count * sizeof (INT16));
outline_list = blob->out_list ();
outline_it.set_to_list (outline_list);
for (i = 0; i < count; i++) {
min[i] = MAX_INT16 - 1;
max[i] = -MAX_INT16 + 1;
y_coord = row_offset + i + 0.5;
for (outline_it.mark_cycle_pt ();
!outline_it.cycled_list (); outline_it.forward ()) {
pts_list = outline_it.data ()->polypts ();
pts_it.set_to_list (pts_list);
for (pts_it.mark_cycle_pt ();
!pts_it.cycled_list (); pts_it.forward ()) {
pt = pts_it.data ()->pos;
vec = pts_it.data ()->vec;
if ((vec.y () != 0) &&
(((pt.y () <= y_coord) && (pt.y () + vec.y () >= y_coord))
|| ((pt.y () >= y_coord)
&& (pt.y () + vec.y () <= y_coord)))) {
/* The segment crosses y_coord so find x-point and check for min/max. */
x_coord = (INT16) floor ((y_coord -
pt.y ()) * vec.x () / vec.y () +
pt.x () + 0.5);
if (x_coord < min[i])
min[i] = x_coord;
x_coord--; //to get pix to left of line
if (x_coord > max[i])
max[i] = x_coord;
}
}
}
}
}
// Computes and returns the displacement of the center of the line
// perpendicular to the given direction.
double BaselineRow::PerpDisp(const FCOORD& direction) const {
float middle_x = (bounding_box_.left() + bounding_box_.right()) / 2.0f;
FCOORD middle_pos(middle_x, StraightYAtX(middle_x));
return direction * middle_pos / direction.length();
}
WordFeature::WordFeature(const FCOORD& fcoord, uinT8 dir)
: x_(IntCastRounded(fcoord.x())),
y_(ClipToRange(IntCastRounded(fcoord.y()), 0, MAX_UINT8)),
dir_(dir) {
}
// Accumulates the segment between pt1 and pt2 in the LLSQ, quantizing over
// the integer coordinate grid to properly weight long vectors.
static void SegmentLLSQ(const FCOORD& pt1, const FCOORD& pt2,
LLSQ* accumulator) {
FCOORD step(pt2);
step -= pt1;
int xstart = IntCastRounded(MIN(pt1.x(), pt2.x()));
int xend = IntCastRounded(MAX(pt1.x(), pt2.x()));
int ystart = IntCastRounded(MIN(pt1.y(), pt2.y()));
int yend = IntCastRounded(MAX(pt1.y(), pt2.y()));
if (xstart == xend && ystart == yend) return; // Nothing to do.
double weight = step.length() / (xend - xstart + yend - ystart);
// Compute and save the y-position at the middle of each x-step.
for (int x = xstart; x < xend; ++x) {
double y = pt1.y() + step.y() * (x + 0.5 - pt1.x()) / step.x();
accumulator->add(x + 0.5, y, weight);
}
// Compute and save the x-position at the middle of each y-step.
for (int y = ystart; y < yend; ++y) {
double x = pt1.x() + step.x() * (y + 0.5 - pt1.y()) / step.y();
accumulator->add(x, y + 0.5, weight);
}
}
void find_blob_limits( //get y limits
PBLOB *blob, //blob to search
float leftx, //x limits
float rightx,
FCOORD rotation, //for landscape
float &ymin, //output y limits
float &ymax) {
float testy; //y intercept
FCOORD pos; //rotated
FCOORD vec;
POLYPT *polypt; //current point
//outlines
OUTLINE_IT out_it = blob->out_list ();
POLYPT_IT poly_it; //outline pts
ymin = (float) MAX_INT32;
ymax = (float) -MAX_INT32;
for (out_it.mark_cycle_pt (); !out_it.cycled_list (); out_it.forward ()) {
//get points
poly_it.set_to_list (out_it.data ()->polypts ());
for (poly_it.mark_cycle_pt (); !poly_it.cycled_list ();
poly_it.forward ()) {
polypt = poly_it.data ();
pos = polypt->pos;
pos.rotate (rotation);
vec = polypt->vec;
vec.rotate (rotation);
if (pos.x () < leftx && pos.x () + vec.x () > leftx
|| pos.x () > leftx && pos.x () + vec.x () < leftx) {
testy = pos.y () + vec.y () * (leftx - pos.x ()) / vec.x ();
//intercept of boundary
if (testy < ymin)
ymin = testy;
if (testy > ymax)
ymax = testy;
}
if (pos.x () >= leftx && pos.x () <= rightx) {
if (pos.y () > ymax)
ymax = pos.y ();
if (pos.y () < ymin)
ymin = pos.y ();
}
if (pos.x () > rightx && pos.x () + vec.x () < rightx
|| pos.x () < rightx && pos.x () + vec.x () > rightx) {
testy = pos.y () + vec.y () * (rightx - pos.x ()) / vec.x ();
//intercept of boundary
if (testy < ymin)
ymin = testy;
if (testy > ymax)
ymax = testy;
}
}
}
}
// Adds any edges from a single segment of outline between pt1 and pt2 to
// the bbox such that it guarantees to contain anything produced by
// SegmentCoords.
static void SegmentBBox(const FCOORD& pt1, const FCOORD& pt2, TBOX* bbox) {
FCOORD step(pt2);
step -= pt1;
int x1 = IntCastRounded(MIN(pt1.x(), pt2.x()));
int x2 = IntCastRounded(MAX(pt1.x(), pt2.x()));
if (x2 > x1) {
int y1 = IntCastRounded(pt1.y() + step.y() * (x1 + 0.5 - pt1.x()) /
step.x());
int y2 = IntCastRounded(pt1.y() + step.y() * (x2 - 0.5 - pt1.x()) /
step.x());
TBOX point(x1, MIN(y1, y2), x2, MAX(y1, y2));
*bbox += point;
}
int y1 = IntCastRounded(MIN(pt1.y(), pt2.y()));
int y2 = IntCastRounded(MAX(pt1.y(), pt2.y()));
if (y2 > y1) {
int x1 = IntCastRounded(pt1.x() + step.x() * (y1 + 0.5 - pt1.y()) /
step.y());
int x2 = IntCastRounded(pt1.x() + step.x() * (y2 - 0.5 - pt1.y()) /
step.y());
TBOX point(MIN(x1, x2), y1, MAX(x1, x2), y2);
*bbox += point;
}
}
void OUTLINE::compute_bb() { //constructor
ICOORD ibl, itr; //integer bb
FCOORD botleft; //bounding box
FCOORD topright;
FCOORD pos; //current pos;
POLYPT_IT polypts = &outline; //iterator
botleft = polypts.data ()->pos;
topright = botleft;
start = ICOORD ((inT16) botleft.x (), (inT16) botleft.y ());
do {
pos = polypts.data ()->pos;
if (pos.x () < botleft.x ())
//get bounding box
botleft = FCOORD (pos.x (), botleft.y ());
if (pos.y () < botleft.y ())
botleft = FCOORD (botleft.x (), pos.y ());
if (pos.x () > topright.x ())
topright = FCOORD (pos.x (), topright.y ());
if (pos.y () > topright.y ())
topright = FCOORD (topright.x (), pos.y ());
polypts.forward ();
}
while (!polypts.at_first ());
ibl = ICOORD ((inT16) botleft.x (), (inT16) botleft.y ());
itr = ICOORD ((inT16) topright.x () + 1, (inT16) topright.y () + 1);
box = TBOX (ibl, itr);
}
// Groups blocks by rotation, then, for each group, makes a WordGrid and calls
// TransferDiacriticsToWords to copy the diacritic blobs to the most
// appropriate words in the group of blocks. Source blobs are not touched.
void Textord::TransferDiacriticsToBlockGroups(BLOBNBOX_LIST* diacritic_blobs,
BLOCK_LIST* blocks) {
// Angle difference larger than this is too much to consider equal.
// They should only be in multiples of M_PI/2 anyway.
const double kMaxAngleDiff = 0.01; // About 0.6 degrees.
PointerVector<BlockGroup> groups;
BLOCK_IT bk_it(blocks);
for (bk_it.mark_cycle_pt(); !bk_it.cycled_list(); bk_it.forward()) {
BLOCK* block = bk_it.data();
if (block->pdblk.poly_block() != nullptr && !block->pdblk.poly_block()->IsText()) {
continue;
}
// Linear search of the groups to find a matching rotation.
float block_angle = block->re_rotation().angle();
int best_g = 0;
float best_angle_diff = MAX_FLOAT32;
for (int g = 0; g < groups.size(); ++g) {
double angle_diff = fabs(block_angle - groups[g]->angle);
if (angle_diff > M_PI) angle_diff = fabs(angle_diff - 2.0 * M_PI);
if (angle_diff < best_angle_diff) {
best_angle_diff = angle_diff;
best_g = g;
}
}
if (best_angle_diff > kMaxAngleDiff) {
groups.push_back(new BlockGroup(block));
} else {
groups[best_g]->blocks.push_back(block);
groups[best_g]->bounding_box += block->pdblk.bounding_box();
float x_height = block->x_height();
if (x_height < groups[best_g]->min_xheight)
groups[best_g]->min_xheight = x_height;
}
}
// Now process each group of blocks.
PointerVector<WordWithBox> word_ptrs;
for (int g = 0; g < groups.size(); ++g) {
const BlockGroup* group = groups[g];
if (group->bounding_box.null_box()) continue;
WordGrid word_grid(group->min_xheight, group->bounding_box.botleft(),
group->bounding_box.topright());
for (int b = 0; b < group->blocks.size(); ++b) {
ROW_IT row_it(group->blocks[b]->row_list());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
ROW* row = row_it.data();
// Put the words of the row into the grid.
WERD_IT w_it(row->word_list());
for (w_it.mark_cycle_pt(); !w_it.cycled_list(); w_it.forward()) {
WERD* word = w_it.data();
WordWithBox* box_word = new WordWithBox(word);
word_grid.InsertBBox(true, true, box_word);
// Save the pointer where it will be auto-deleted.
word_ptrs.push_back(box_word);
}
}
}
FCOORD rotation = group->rotation;
// Make it a forward rotation that will transform blob coords to block.
rotation.set_y(-rotation.y());
TransferDiacriticsToWords(diacritic_blobs, rotation, &word_grid);
}
}
// Returns the direction of the fitted line as a unit vector, using the
// least mean squared perpendicular distance. The line runs through the
// mean_point, i.e. a point p on the line is given by:
// p = mean_point() + lambda * vector_fit() for some real number lambda.
// Note that the result (0<=x<=1, -1<=y<=-1) is directionally ambiguous
// and may be negated without changing its meaning.
FCOORD LLSQ::vector_fit() const {
double x_var = x_variance();
double y_var = y_variance();
double covar = covariance();
FCOORD result;
if (x_var >= y_var) {
if (x_var == 0.0)
return FCOORD(0.0f, 0.0f);
result.set_x(x_var / sqrt(x_var * x_var + covar * covar));
result.set_y(sqrt(1.0 - result.x() * result.x()));
} else {
result.set_y(y_var / sqrt(y_var * y_var + covar * covar));
result.set_x(sqrt(1.0 - result.y() * result.y()));
}
if (covar < 0.0)
result.set_y(-result.y());
return result;
}
