// Draws the outline in the given colour, normalized using the given denorm,
// making use of sub-pixel accurate information if available.
void C_OUTLINE::plot_normed(const DENORM& denorm, ScrollView::Color colour,
ScrollView* window) const {
window->Pen(colour);
if (stepcount == 0) {
window->Rectangle(box.left(), box.top(), box.right(), box.bottom());
return;
}
const DENORM* root_denorm = denorm.RootDenorm();
ICOORD pos = start; // current position
FCOORD f_pos = sub_pixel_pos_at_index(pos, 0);
FCOORD pos_normed;
denorm.NormTransform(root_denorm, f_pos, &pos_normed);
window->SetCursor(IntCastRounded(pos_normed.x()),
IntCastRounded(pos_normed.y()));
for (int s = 0; s < stepcount; pos += step(s++)) {
int edge_weight = edge_strength_at_index(s);
if (edge_weight == 0) {
// This point has conflicting gradient and step direction, so ignore it.
continue;
}
FCOORD f_pos = sub_pixel_pos_at_index(pos, s);
FCOORD pos_normed;
denorm.NormTransform(root_denorm, f_pos, &pos_normed);
window->DrawTo(IntCastRounded(pos_normed.x()),
IntCastRounded(pos_normed.y()));
}
}
// Normalize in-place using the DENORM.
void TESSLINE::Normalize(const DENORM& denorm) {
EDGEPT* pt = loop;
do {
denorm.LocalNormTransform(pt->pos, &pt->pos);
pt = pt->next;
} while (pt != loop);
SetupFromPos();
}
// Normalize in-place using the DENORM.
void TBLOB::Normalize(const DENORM& denorm) {
// TODO(rays) outline->Normalize is more accurate, but breaks tests due
// the changes it makes. Reinstate this code with a retraining.
#if 1
for (TESSLINE* outline = outlines; outline != NULL; outline = outline->next) {
outline->Normalize(denorm);
}
#else
denorm.LocalNormBlob(this);
#endif
}
// Helper normalizes the direction, assuming that it is at the given
// unnormed_pos, using the given denorm, starting at the root_denorm.
uint8_t NormalizeDirection(uint8_t dir, const FCOORD& unnormed_pos,
const DENORM& denorm, const DENORM* root_denorm) {
// Convert direction to a vector.
FCOORD unnormed_end;
unnormed_end.from_direction(dir);
unnormed_end += unnormed_pos;
FCOORD normed_pos, normed_end;
denorm.NormTransform(root_denorm, unnormed_pos, &normed_pos);
denorm.NormTransform(root_denorm, unnormed_end, &normed_end);
normed_end -= normed_pos;
return normed_end.to_direction();
}
BLOB_CHOICE* M_Utils::runBlobOCR(BLOBNBOX* blob, Tesseract* ocrengine) {
// * Normalize blob height to x-height (current OSD):
// SetupNormalization(NULL, NULL, &rotation, NULL, NULL, 0,
// box.rotational_x_middle(rotation),
// box.rotational_y_middle(rotation),
// kBlnXHeight / box.rotational_height(rotation),
// kBlnXHeight / box.rotational_height(rotation),
// 0, kBlnBaselineOffset);
BLOB_CHOICE_LIST ratings_lang;
C_BLOB* cblob = blob->cblob();
TBLOB* tblob = TBLOB::PolygonalCopy(cblob);
const TBOX& box = tblob->bounding_box();
// Normalize the blob. Set the origin to the place we want to be the
// bottom-middle, and scaling is to mpx, box_, NULL);
float scaling = static_cast<float>(kBlnXHeight) / box.height();
DENORM denorm;
float x_orig = (box.left() + box.right()) / 2.0f, y_orig = box.bottom();
denorm.SetupNormalization(NULL, NULL, NULL, NULL, NULL, 0,
x_orig, y_orig, scaling, scaling,
0.0f, static_cast<float>(kBlnBaselineOffset));
TBLOB* normed_blob = new TBLOB(*tblob);
normed_blob->Normalize(denorm);
ocrengine->AdaptiveClassifier(normed_blob, denorm, &ratings_lang, NULL);
delete normed_blob;
delete tblob;
// Get the best choice from ratings_lang and rating_equ. As the choice in the
// list has already been sorted by the certainty, we simply use the first
// choice.
BLOB_CHOICE *lang_choice = NULL;
if (ratings_lang.length() > 0) {
BLOB_CHOICE_IT choice_it(&ratings_lang);
lang_choice = choice_it.data();
}
return lang_choice;
}
// 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;
}
// Normalizes the blob for classification only if needed.
// (Normally this means a non-zero classify rotation.)
// If no Normalization is needed, then NULL is returned, and the denorm is
// unchanged. Otherwise a new TBLOB is returned and the denorm points to
// a new DENORM. In this case, both the TBLOB and DENORM must be deleted.
TBLOB* TBLOB::ClassifyNormalizeIfNeeded(const DENORM** denorm) const {
TBLOB* rotated_blob = NULL;
// If necessary, copy the blob and rotate it. The rotation is always
// +/- 90 degrees, as 180 was already taken care of.
if ((*denorm)->block() != NULL &&
(*denorm)->block()->classify_rotation().y() != 0.0) {
TBOX box = bounding_box();
int x_middle = (box.left() + box.right()) / 2;
int y_middle = (box.top() + box.bottom()) / 2;
rotated_blob = new TBLOB(*this);
const FCOORD& rotation = (*denorm)->block()->classify_rotation();
DENORM* norm = new DENORM;
// Move the rotated blob back to the same y-position so that we
// can still distinguish similar glyphs with differeny y-position.
float target_y = kBlnBaselineOffset +
(rotation.y() > 0 ? x_middle - box.left() : box.right() - x_middle);
norm->SetupNormalization(NULL, NULL, &rotation, *denorm, NULL, 0,
x_middle, y_middle, 1.0f, 1.0f, 0.0f, target_y);
// x_middle, y_middle, 1.0f, 1.0f, 0.0f, y_middle);
rotated_blob->Normalize(*norm);
*denorm = norm;
}
return rotated_blob;
}
// Extracts Tesseract features and appends them to the features vector.
// Startpt to lastpt, inclusive, MUST have the same src_outline member,
// which may be nullptr. The vector from lastpt to its next is included in
// the feature extraction. Hidden edges should be excluded by the caller.
// If force_poly is true, the features will be extracted from the polygonal
// approximation even if more accurate data is available.
static void ExtractFeaturesFromRun(
const EDGEPT* startpt, const EDGEPT* lastpt,
const DENORM& denorm, double feature_length, bool force_poly,
GenericVector<INT_FEATURE_STRUCT>* features) {
const EDGEPT* endpt = lastpt->next;
const C_OUTLINE* outline = startpt->src_outline;
if (outline != nullptr && !force_poly) {
// Detailed information is available. We have to normalize only from
// the root_denorm to denorm.
const DENORM* root_denorm = denorm.RootDenorm();
int total_features = 0;
// Get the features from the outline.
int step_length = outline->pathlength();
int start_index = startpt->start_step;
// pos is the integer coordinates of the binary image steps.
ICOORD pos = outline->position_at_index(start_index);
// We use an end_index that allows us to use a positive increment, but that
// may be beyond the bounds of the outline steps/ due to wrap-around, to
// so we use % step_length everywhere, except for start_index.
int end_index = lastpt->start_step + lastpt->step_count;
if (end_index <= start_index)
end_index += step_length;
LLSQ prev_points;
LLSQ prev_dirs;
FCOORD prev_normed_pos = outline->sub_pixel_pos_at_index(pos, start_index);
denorm.NormTransform(root_denorm, prev_normed_pos, &prev_normed_pos);
LLSQ points;
LLSQ dirs;
FCOORD normed_pos(0.0f, 0.0f);
int index = GatherPoints(outline, feature_length, denorm, root_denorm,
start_index, end_index, &pos, &normed_pos,
&points, &dirs);
while (index <= end_index) {
// At each iteration we nominally have 3 accumulated sets of points and
// dirs: prev_points/dirs, points/dirs, next_points/dirs and sum them
// into sum_points/dirs, but we don't necessarily get any features out,
// so if that is the case, we keep accumulating instead of rotating the
// accumulators.
LLSQ next_points;
LLSQ next_dirs;
FCOORD next_normed_pos(0.0f, 0.0f);
index = GatherPoints(outline, feature_length, denorm, root_denorm,
index, end_index, &pos, &next_normed_pos,
&next_points, &next_dirs);
LLSQ sum_points(prev_points);
// TODO(rays) find out why it is better to use just dirs and next_dirs
// in sum_dirs, instead of using prev_dirs as well.
LLSQ sum_dirs(dirs);
sum_points.add(points);
sum_points.add(next_points);
sum_dirs.add(next_dirs);
bool made_features = false;
// If we have some points, we can try making some features.
if (sum_points.count() > 0) {
// We have gone far enough from the start. Make a feature and restart.
FCOORD fit_pt = sum_points.mean_point();
FCOORD fit_vector = MeanDirectionVector(sum_points, sum_dirs,
prev_normed_pos, normed_pos);
// The segment to which we fit features is the line passing through
// fit_pt in direction of fit_vector that starts nearest to
// prev_normed_pos and ends nearest to normed_pos.
FCOORD start_pos = prev_normed_pos.nearest_pt_on_line(fit_pt,
fit_vector);
FCOORD end_pos = normed_pos.nearest_pt_on_line(fit_pt, fit_vector);
// Possible correction to match the adjacent polygon segment.
if (total_features == 0 && startpt != endpt) {
FCOORD poly_pos(startpt->pos.x, startpt->pos.y);
denorm.LocalNormTransform(poly_pos, &start_pos);
}
if (index > end_index && startpt != endpt) {
FCOORD poly_pos(endpt->pos.x, endpt->pos.y);
denorm.LocalNormTransform(poly_pos, &end_pos);
}
int num_features = ComputeFeatures(start_pos, end_pos, feature_length,
features);
if (num_features > 0) {
// We made some features so shuffle the accumulators.
prev_points = points;
prev_dirs = dirs;
prev_normed_pos = normed_pos;
points = next_points;
dirs = next_dirs;
made_features = true;
total_features += num_features;
}
// The end of the next set becomes the end next time around.
normed_pos = next_normed_pos;
}
if (!made_features) {
// We didn't make any features, so keep the prev accumulators and
// add the next ones into the current.
points.add(next_points);
dirs.add(next_dirs);
}
}
} else {
// There is no outline, so we are forced to use the polygonal approximation.
const EDGEPT* pt = startpt;
do {
FCOORD start_pos(pt->pos.x, pt->pos.y);
FCOORD end_pos(pt->next->pos.x, pt->next->pos.y);
denorm.LocalNormTransform(start_pos, &start_pos);
denorm.LocalNormTransform(end_pos, &end_pos);
ComputeFeatures(start_pos, end_pos, feature_length, features);
} while ((pt = pt->next) != endpt);
}
}
// Collects edges into the given bounding box, LLSQ accumulator and/or x_coords,
// y_coords vectors.
// For a description of x_coords/y_coords, see GetEdgeCoords above.
// Startpt to lastpt, inclusive, MUST have the same src_outline member,
// which may be NULL. The vector from lastpt to its next is included in
// the accumulation. Hidden edges should be excluded by the caller.
// The input denorm should be the normalizations that have been applied from
// the image to the current state of the TBLOB from which startpt, lastpt come.
// box is the bounding box of the blob from which the EDGEPTs are taken and
// indices into x_coords, y_coords are offset by box.botleft().
static void CollectEdgesOfRun(const EDGEPT* startpt, const EDGEPT* lastpt,
const DENORM& denorm, const TBOX& box,
TBOX* bounding_box,
LLSQ* accumulator,
GenericVector<GenericVector<int> > *x_coords,
GenericVector<GenericVector<int> > *y_coords) {
const C_OUTLINE* outline = startpt->src_outline;
int x_limit = box.width() - 1;
int y_limit = box.height() - 1;
if (outline != NULL) {
// Use higher-resolution edge points stored on the outline.
// The outline coordinates may not match the binary image because of the
// rotation for vertical text lines, but the root_denorm IS the matching
// start of the DENORM chain.
const DENORM* root_denorm = denorm.RootDenorm();
int step_length = outline->pathlength();
int start_index = startpt->start_step;
// Note that if this run straddles the wrap-around point of the outline,
// that lastpt->start_step may have a lower index than startpt->start_step,
// and we want to use an end_index that allows us to use a positive
// increment, so we add step_length if necessary, but that may be beyond the
// bounds of the outline steps/ due to wrap-around, so we use % step_length
// everywhere, except for start_index.
int end_index = lastpt->start_step + lastpt->step_count;
if (end_index <= start_index)
end_index += step_length;
// pos is the integer coordinates of the binary image steps.
ICOORD pos = outline->position_at_index(start_index);
FCOORD origin(box.left(), box.bottom());
// f_pos is a floating-point version of pos that offers improved edge
// positioning using greyscale information or smoothing of edge steps.
FCOORD f_pos = outline->sub_pixel_pos_at_index(pos, start_index);
// pos_normed is f_pos after the appropriate normalization, and relative
// to origin.
// prev_normed is the previous value of pos_normed.
FCOORD prev_normed;
denorm.NormTransform(root_denorm, f_pos, &prev_normed);
prev_normed -= origin;
for (int index = start_index; index < end_index; ++index) {
ICOORD step = outline->step(index % step_length);
// Only use the point if its edge strength is positive. This excludes
// points that don't provide useful information, eg
// ___________
// |___________
// The vertical step provides only noisy, damaging information, as even
// with a greyscale image, the positioning of the edge there may be a
// fictitious extrapolation, so previous processing has eliminated it.
if (outline->edge_strength_at_index(index % step_length) > 0) {
FCOORD f_pos = outline->sub_pixel_pos_at_index(pos,
index % step_length);
FCOORD pos_normed;
denorm.NormTransform(root_denorm, f_pos, &pos_normed);
pos_normed -= origin;
// Accumulate the information that is selected by the caller.
if (bounding_box != NULL) {
SegmentBBox(pos_normed, prev_normed, bounding_box);
}
if (accumulator != NULL) {
SegmentLLSQ(pos_normed, prev_normed, accumulator);
}
if (x_coords != NULL && y_coords != NULL) {
SegmentCoords(pos_normed, prev_normed, x_limit, y_limit,
x_coords, y_coords);
}
prev_normed = pos_normed;
}
pos += step;
}
} else {
// There is no outline, so we are forced to use the polygonal approximation.
const EDGEPT* endpt = lastpt->next;
const EDGEPT* pt = startpt;
do {
FCOORD next_pos(pt->next->pos.x - box.left(),
pt->next->pos.y - box.bottom());
FCOORD pos(pt->pos.x - box.left(), pt->pos.y - box.bottom());
if (bounding_box != NULL) {
SegmentBBox(next_pos, pos, bounding_box);
}
if (accumulator != NULL) {
SegmentLLSQ(next_pos, pos, accumulator);
}
if (x_coords != NULL && y_coords != NULL) {
SegmentCoords(next_pos, pos, x_limit, y_limit, x_coords, y_coords);
}
} while ((pt = pt->next) != endpt);
}
}