void BLOCK::compress() { // squash it up
#define ROW_SPACING 5
ROW_IT row_it(&rows);
ROW *row;
ICOORD row_spacing (0, ROW_SPACING);
ICOORDELT_IT icoordelt_it;
sort_rows();
box = TBOX (box.topleft (), box.topleft ());
box.move_bottom_edge (ROW_SPACING);
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
row->move (box.botleft () - row_spacing -
row->bounding_box ().topleft ());
box += row->bounding_box ();
}
leftside.clear ();
icoordelt_it.set_to_list (&leftside);
icoordelt_it.add_to_end (new ICOORDELT (box.left (), box.bottom ()));
icoordelt_it.add_to_end (new ICOORDELT (box.left (), box.top ()));
rightside.clear ();
icoordelt_it.set_to_list (&rightside);
icoordelt_it.add_to_end (new ICOORDELT (box.right (), box.bottom ()));
icoordelt_it.add_to_end (new ICOORDELT (box.right (), box.top ()));
}
void PrintSegmentationStats(BLOCK_LIST* block_list) {
int num_blocks = 0;
int num_rows = 0;
int num_words = 0;
int num_blobs = 0;
BLOCK_IT block_it(block_list);
for (block_it.mark_cycle_pt(); !block_it.cycled_list(); block_it.forward()) {
BLOCK* block = block_it.data();
++num_blocks;
ROW_IT row_it(block->row_list());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
++num_rows;
ROW* row = row_it.data();
// Iterate over all werds in the row.
WERD_IT werd_it(row->word_list());
for (werd_it.mark_cycle_pt(); !werd_it.cycled_list(); werd_it.forward()) {
WERD* werd = werd_it.data();
++num_words;
num_blobs += werd->cblob_list()->length();
}
}
}
tprintf("Block list stats:\nBlocks = %d\nRows = %d\nWords = %d\nBlobs = %d\n",
num_blocks, num_rows, num_words, num_blobs);
}
void make_single_word(bool one_blob, TO_ROW_LIST *rows, ROW_LIST* real_rows) {
TO_ROW_IT to_row_it(rows);
TO_ROW* row = to_row_it.data();
// The blobs have to come out of the BLOBNBOX into the C_BLOB_LIST ready
// to create the word.
C_BLOB_LIST cblobs;
C_BLOB_IT cblob_it(&cblobs);
BLOBNBOX_IT box_it(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(row, static_cast<inT16>(row->kern_size),
static_cast<inT16>(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_it.add_after_then_move(word);
ROW_IT row_it(real_rows);
row_it.add_after_then_move(real_row);
}
//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;
}
BaselineBlock::BaselineBlock(int debug_level, bool non_text, TO_BLOCK* block)
: block_(block), debug_level_(debug_level), non_text_block_(non_text),
good_skew_angle_(false), skew_angle_(0.0),
line_spacing_(block->line_spacing), line_offset_(0.0), model_error_(0.0) {
TO_ROW_IT row_it(block_->get_rows());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
// Sort the blobs on the rows.
row_it.data()->blob_list()->sort(blob_x_order);
rows_.push_back(new BaselineRow(block->line_spacing, row_it.data()));
}
}
// Helper computes median xheight in the image.
static double MedianXHeight(BLOCK_LIST *block_list) {
BLOCK_IT block_it(block_list);
STATS xheights(0, block_it.data()->bounding_box().height());
for (block_it.mark_cycle_pt();
!block_it.cycled_list(); block_it.forward()) {
ROW_IT row_it(block_it.data()->row_list());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
xheights.add(IntCastRounded(row_it.data()->x_height()), 1);
}
}
return xheights.median();
}
static void clear_any_old_text(BLOCK_LIST *block_list) {
BLOCK_IT block_it(block_list);
for (block_it.mark_cycle_pt();
!block_it.cycled_list(); block_it.forward()) {
ROW_IT row_it(block_it.data()->row_list());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
WERD_IT word_it(row_it.data()->word_list());
for (word_it.mark_cycle_pt();
!word_it.cycled_list(); word_it.forward()) {
word_it.data()->set_text("");
}
}
}
}
// Sets the parameters in TO_BLOCK that are needed by subsequent processes.
void BaselineBlock::SetupBlockParameters() const {
if (line_spacing_ > 0.0) {
// Where was block_line_spacing set before?
float min_spacing = MIN(block_->line_spacing, line_spacing_);
if (min_spacing < block_->line_size)
block_->line_size = min_spacing;
block_->line_spacing = line_spacing_;
block_->baseline_offset = line_offset_;
block_->max_blob_size = line_spacing_ * kMaxBlobSizeMultiple;
}
// Setup the parameters on all the rows.
TO_ROW_IT row_it(block_->get_rows());
for (int r = 0; r < rows_.size(); ++r, row_it.forward()) {
BaselineRow* row = rows_[r];
TO_ROW* to_row = row_it.data();
row->SetupOldLineParameters(to_row);
}
}
// This method resolves the cc bbox to a particular row and returns the row's
// xheight.
int ShiroRekhaSplitter::GetXheightForCC(Box* cc_bbox) {
if (!segmentation_block_list_) {
return global_xheight_;
}
// Compute the box coordinates in Tesseract's coordinate system.
TBOX bbox(cc_bbox->x,
pixGetHeight(orig_pix_) - cc_bbox->y - cc_bbox->h - 1,
cc_bbox->x + cc_bbox->w,
pixGetHeight(orig_pix_) - cc_bbox->y - 1);
// Iterate over all blocks.
BLOCK_IT block_it(segmentation_block_list_);
for (block_it.mark_cycle_pt(); !block_it.cycled_list(); block_it.forward()) {
BLOCK* block = block_it.data();
// Iterate over all rows in the block.
ROW_IT row_it(block->row_list());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
ROW* row = row_it.data();
if (!row->bounding_box().major_overlap(bbox)) {
continue;
}
// Row could be skewed, warped, etc. Use the position of the box to
// determine the baseline position of the row for that x-coordinate.
// Create a square TBOX whose baseline's mid-point lies at this point
// and side is row's xheight. Take the overlap of this box with the input
// box and check if it is a 'major overlap'. If so, this box lies in this
// row. In that case, return the xheight for this row.
float box_middle = 0.5 * (bbox.left() + bbox.right());
int baseline = static_cast<int>(row->base_line(box_middle) + 0.5);
TBOX test_box(box_middle - row->x_height() / 2,
baseline,
box_middle + row->x_height() / 2,
static_cast<int>(baseline + row->x_height()));
// Compute overlap. If it is is a major overlap, this is the right row.
if (bbox.major_overlap(test_box)) {
return row->x_height();
}
}
}
// No row found for this bbox.
return kUnspecifiedXheight;
}
void RefreshWordBlobsFromNewBlobs(BLOCK_LIST* block_list,
C_BLOB_LIST* new_blobs,
C_BLOB_LIST* not_found_blobs) {
// Now iterate over all the blobs in the segmentation_block_list_, and just
// replace the corresponding c-blobs inside the werds.
BLOCK_IT block_it(block_list);
for (block_it.mark_cycle_pt(); !block_it.cycled_list(); block_it.forward()) {
BLOCK* block = block_it.data();
if (block->poly_block() != NULL && !block->poly_block()->IsText())
continue; // Don't touch non-text blocks.
// Iterate over all rows in the block.
ROW_IT row_it(block->row_list());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
ROW* row = row_it.data();
// Iterate over all werds in the row.
WERD_IT werd_it(row->word_list());
WERD_LIST new_words;
WERD_IT new_words_it(&new_words);
for (werd_it.mark_cycle_pt(); !werd_it.cycled_list(); werd_it.forward()) {
WERD* werd = werd_it.extract();
WERD* new_werd = werd->ConstructWerdWithNewBlobs(new_blobs,
not_found_blobs);
if (new_werd) {
// Insert this new werd into the actual row's werd-list. Remove the
// existing one.
new_words_it.add_after_then_move(new_werd);
delete werd;
} else {
// Reinsert the older word back, for lack of better options.
// This is critical since dropping the words messes up segmentation:
// eg. 1st word in the row might otherwise have W_FUZZY_NON turned on.
new_words_it.add_after_then_move(werd);
}
}
// Get rid of the old word list & replace it with the new one.
row->word_list()->clear();
werd_it.move_to_first();
werd_it.add_list_after(&new_words);
}
}
}
void ExtractBlobsFromSegmentation(BLOCK_LIST* blocks,
C_BLOB_LIST* output_blob_list) {
C_BLOB_IT return_list_it(output_blob_list);
BLOCK_IT block_it(blocks);
for (block_it.mark_cycle_pt(); !block_it.cycled_list(); block_it.forward()) {
BLOCK* block = block_it.data();
ROW_IT row_it(block->row_list());
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
ROW* row = row_it.data();
// Iterate over all werds in the row.
WERD_IT werd_it(row->word_list());
for (werd_it.mark_cycle_pt(); !werd_it.cycled_list(); werd_it.forward()) {
WERD* werd = werd_it.data();
return_list_it.move_to_last();
return_list_it.add_list_after(werd->cblob_list());
return_list_it.move_to_last();
return_list_it.add_list_after(werd->rej_cblob_list());
}
}
}
}
// Fixes the block so it obeys all the rules:
// Must have at least one ROW.
// Must have at least one WERD.
// WERDs contain a fake blob.
void Textord::cleanup_nontext_block(BLOCK* block) {
// Non-text blocks must contain at least one row.
ROW_IT row_it(block->row_list());
if (row_it.empty()) {
const TBOX& box = block->pdblk.bounding_box();
float height = box.height();
int32_t xstarts[2] = {box.left(), box.right()};
double coeffs[3] = {0.0, 0.0, static_cast<double>(box.bottom())};
ROW* row = new ROW(1, xstarts, coeffs, height / 2.0f, height / 4.0f,
height / 4.0f, 0, 1);
row_it.add_after_then_move(row);
}
// Each row must contain at least one word.
for (row_it.mark_cycle_pt(); !row_it.cycled_list(); row_it.forward()) {
ROW* row = row_it.data();
WERD_IT w_it(row->word_list());
if (w_it.empty()) {
// Make a fake blob to put in the word.
TBOX box = block->row_list()->singleton() ? block->pdblk.bounding_box()
: row->bounding_box();
C_BLOB* blob = C_BLOB::FakeBlob(box);
C_BLOB_LIST blobs;
C_BLOB_IT blob_it(&blobs);
blob_it.add_after_then_move(blob);
WERD* word = new WERD(&blobs, 0, nullptr);
w_it.add_after_then_move(word);
}
// Each word must contain a fake blob.
for (w_it.mark_cycle_pt(); !w_it.cycled_list(); w_it.forward()) {
WERD* word = w_it.data();
// Just assert that this is true, as it would be useful to find
// out why it isn't.
ASSERT_HOST(!word->cblob_list()->empty());
}
row->recalc_bounding_box();
}
}
void BLOCK::sort_rows() { // order on "top"
ROW_IT row_it(&rows);
row_it.sort (decreasing_top_order);
}
// 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);
}
}
GAPMAP::GAPMAP( //Constructor
TO_BLOCK *block //block
) {
TO_ROW *row; //current row
BLOBNBOX_IT blob_it; //iterator
TBOX blob_box;
TBOX prev_blob_box;
int16_t gap_width;
int16_t start_of_row;
int16_t end_of_row;
STATS xht_stats (0, 128);
int16_t min_quantum;
int16_t max_quantum;
int16_t i;
/*
Find left and right extremes and bucket size
*/
map = nullptr;
min_left = INT16_MAX;
max_right = -INT16_MAX;
total_rows = 0;
any_tabs = false;
// row iterator
TO_ROW_IT row_it(block->get_rows());
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
if (!row->blob_list ()->empty ()) {
total_rows++;
xht_stats.add (static_cast<int16_t>(floor (row->xheight + 0.5)), 1);
blob_it.set_to_list (row->blob_list ());
start_of_row = blob_it.data ()->bounding_box ().left ();
end_of_row = blob_it.data_relative (-1)->bounding_box ().right ();
if (min_left > start_of_row)
min_left = start_of_row;
if (max_right < end_of_row)
max_right = end_of_row;
}
}
if ((total_rows < 3) || (min_left >= max_right)) {
bucket_size = 0;
map_max = 0;
total_rows = 0;
min_left = max_right = 0;
return;
}
bucket_size = static_cast<int16_t>(floor (xht_stats.median () + 0.5)) / 2;
map_max = (max_right - min_left) / bucket_size;
map = new int16_t[map_max + 1];
for (i = 0; i <= map_max; i++)
map[i] = 0;
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
if (!row->blob_list ()->empty ()) {
blob_it.set_to_list (row->blob_list ());
blob_it.mark_cycle_pt ();
blob_box = box_next (&blob_it);
prev_blob_box = blob_box;
if (gapmap_use_ends) {
/* Leading space */
gap_width = blob_box.left () - min_left;
if ((gap_width > gapmap_big_gaps * row->xheight)
&& gap_width > 2) {
max_quantum = (blob_box.left () - min_left) / bucket_size;
if (max_quantum > map_max) max_quantum = map_max;
for (i = 0; i <= max_quantum; i++)
map[i]++;
}
}
while (!blob_it.cycled_list ()) {
blob_box = box_next (&blob_it);
gap_width = blob_box.left () - prev_blob_box.right ();
if ((gap_width > gapmap_big_gaps * row->xheight)
&& gap_width > 2) {
min_quantum =
(prev_blob_box.right () - min_left) / bucket_size;
max_quantum = (blob_box.left () - min_left) / bucket_size;
if (max_quantum > map_max) max_quantum = map_max;
for (i = min_quantum; i <= max_quantum; i++)
map[i]++;
}
prev_blob_box = blob_box;
}
if (gapmap_use_ends) {
/* Trailing space */
gap_width = max_right - prev_blob_box.right ();
if ((gap_width > gapmap_big_gaps * row->xheight)
&& gap_width > 2) {
min_quantum =
(prev_blob_box.right () - min_left) / bucket_size;
if (min_quantum < 0) min_quantum = 0;
for (i = min_quantum; i <= map_max; i++)
map[i]++;
}
}
}
}
for (i = 0; i <= map_max; i++) {
if (map[i] > total_rows / 2) {
if (gapmap_no_isolated_quanta &&
(((i == 0) &&
(map[i + 1] <= total_rows / 2)) ||
((i == map_max) &&
(map[i - 1] <= total_rows / 2)) ||
((i > 0) &&
(i < map_max) &&
(map[i - 1] <= total_rows / 2) &&
(map[i + 1] <= total_rows / 2)))) {
map[i] = 0; //prevent isolated quantum
}
else
any_tabs = true;
}
}
if (gapmap_debug && any_tabs)
tprintf ("Table found\n");
}