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);
}
WERD *add_repeated_word( //move repeated word
WERD_IT *rep_it, //repeated words
int16_t &rep_left, //left edge of word
int16_t &prev_chop_coord, //previous word end
uint8_t &blanks, //no of blanks
float pitch, //char cell size
WERD_IT *word_it //list of words
) {
WERD *word; //word to move
int16_t new_blanks; //extra blanks
if (rep_left > prev_chop_coord) {
new_blanks = (uint8_t) floor ((rep_left - prev_chop_coord) / pitch + 0.5);
blanks += new_blanks;
}
word = rep_it->extract ();
prev_chop_coord = word->bounding_box ().right ();
word_it->add_after_then_move (word);
word->set_blanks (blanks);
rep_it->forward ();
if (rep_it->empty ())
rep_left = INT16_MAX;
else
rep_left = rep_it->data ()->bounding_box ().left ();
blanks = 0;
return word;
}
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);
}
PAGE_RES_IT* make_pseudo_word(PAGE_RES* page_res, const TBOX& selection_box) {
PAGE_RES_IT pr_it(page_res);
C_BLOB_LIST new_blobs; // list of gathered blobs
C_BLOB_IT new_blob_it = &new_blobs; // iterator
for (WERD_RES* word_res = pr_it.word(); word_res != NULL;
word_res = pr_it.forward()) {
WERD* word = word_res->word;
if (word->bounding_box().overlap(selection_box)) {
C_BLOB_IT blob_it(word->cblob_list());
for (blob_it.mark_cycle_pt();
!blob_it.cycled_list(); blob_it.forward()) {
C_BLOB* blob = blob_it.data();
if (blob->bounding_box().overlap(selection_box)) {
new_blob_it.add_after_then_move(C_BLOB::deep_copy(blob));
}
}
if (!new_blobs.empty()) {
WERD* pseudo_word = new WERD(&new_blobs, 1, NULL);
word_res = pr_it.InsertSimpleCloneWord(*word_res, pseudo_word);
PAGE_RES_IT* it = new PAGE_RES_IT(page_res);
while (it->word() != word_res && it->word() != NULL) it->forward();
ASSERT_HOST(it->word() == word_res);
return it;
}
}
}
return NULL;
}
/// Builds a PAGE_RES from the block_list in the way required for ApplyBoxes:
/// All fuzzy spaces are removed, and all the words are maximally chopped.
PAGE_RES* Tesseract::SetupApplyBoxes(const GenericVector<TBOX>& boxes,
BLOCK_LIST *block_list) {
PreenXHeights(block_list);
// Strip all fuzzy space markers to simplify the PAGE_RES.
BLOCK_IT b_it(block_list);
for (b_it.mark_cycle_pt(); !b_it.cycled_list(); b_it.forward()) {
BLOCK* block = b_it.data();
ROW_IT r_it(block->row_list());
for (r_it.mark_cycle_pt(); !r_it.cycled_list(); r_it.forward ()) {
ROW* row = r_it.data();
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();
if (word->cblob_list()->empty()) {
delete w_it.extract();
} else {
word->set_flag(W_FUZZY_SP, false);
word->set_flag(W_FUZZY_NON, false);
}
}
}
}
PAGE_RES* page_res = new PAGE_RES(false, block_list, NULL);
PAGE_RES_IT pr_it(page_res);
WERD_RES* word_res;
while ((word_res = pr_it.word()) != NULL) {
MaximallyChopWord(boxes, pr_it.block()->block,
pr_it.row()->row, word_res);
pr_it.forward();
}
return page_res;
}
// Remove outlines that are a tiny fraction in either width or height
// of the word height.
void Textord::clean_small_noise_from_words(ROW *row) {
WERD_IT word_it(row->word_list());
for (word_it.mark_cycle_pt(); !word_it.cycled_list(); word_it.forward()) {
WERD* word = word_it.data();
int min_size = static_cast<int>(
textord_noise_hfract * word->bounding_box().height() + 0.5);
C_BLOB_IT blob_it(word->cblob_list());
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
C_BLOB* blob = blob_it.data();
C_OUTLINE_IT out_it(blob->out_list());
for (out_it.mark_cycle_pt(); !out_it.cycled_list(); out_it.forward()) {
C_OUTLINE* outline = out_it.data();
outline->RemoveSmallRecursive(min_size, &out_it);
}
if (blob->out_list()->empty()) {
delete blob_it.extract();
}
}
if (word->cblob_list()->empty()) {
if (!word_it.at_last()) {
// The next word is no longer a fuzzy non space if it was before,
// since the word before is about to be deleted.
WERD* next_word = word_it.data_relative(1);
if (next_word->flag(W_FUZZY_NON)) {
next_word->set_flag(W_FUZZY_NON, false);
}
}
delete word_it.extract();
}
}
}
//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;
}
WERD *make_real_word(BLOBNBOX_IT *box_it, //iterator
inT32 blobcount, //no of blobs to use
BOOL8 bol, //start of line
uinT8 blanks //no of blanks
) {
OUTLINE_IT out_it; // outlines
C_OUTLINE_IT cout_it;
PBLOB_LIST blobs; // blobs in word
C_BLOB_LIST cblobs;
PBLOB_IT blob_it = &blobs; // iterator
C_BLOB_IT cblob_it = &cblobs;
WERD *word; // new word
BLOBNBOX *bblob; // current blob
inT32 blobindex; // in row
for (blobindex = 0; blobindex < blobcount; blobindex++) {
bblob = box_it->extract();
if (bblob->joined_to_prev()) {
if (bblob->blob() != NULL) {
out_it.set_to_list(blob_it.data()->out_list());
out_it.move_to_last();
out_it.add_list_after(bblob->blob()->out_list());
delete bblob->blob();
}
else if (bblob->cblob() != NULL) {
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->blob() != NULL)
blob_it.add_after_then_move(bblob->blob());
else if (bblob->cblob() != NULL)
cblob_it.add_after_then_move(bblob->cblob());
}
delete bblob;
box_it->forward(); // next one
}
if (blanks < 1)
blanks = 1;
if (blob_it.empty())
word = new WERD(&cblobs, blanks, NULL);
else
word = new WERD(&blobs, blanks, NULL);
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;
}
void ROW::plot( //draw it
ScrollView* window //window to draw in
) {
WERD *word; //current word
WERD_IT it = &words; //words of ROW
for (it.mark_cycle_pt (); !it.cycled_list (); it.forward ()) {
word = it.data ();
word->plot (window); //in rainbow colours
}
}
void ROW::plot( //draw it
ScrollView* window, //window to draw in
ScrollView::Color colour //colour to draw in
) {
WERD *word; //current word
WERD_IT it = &words; //words of ROW
for (it.mark_cycle_pt (); !it.cycled_list (); it.forward ()) {
word = it.data ();
word->plot (window, colour); //all in one colour
}
}
/**
* @name process_selected_words()
*
* Walk the current block list applying the specified word processor function
* to each word that overlaps the selection_box.
*/
void Tesseract::process_selected_words(
PAGE_RES* page_res, // blocks to check
TBOX & selection_box,
BOOL8(tesseract::Tesseract::*word_processor)(PAGE_RES_IT* pr_it)) {
for (PAGE_RES_IT page_res_it(page_res); page_res_it.word() != NULL;
page_res_it.forward()) {
WERD* word = page_res_it.word()->word;
if (word->bounding_box().overlap(selection_box)) {
if (!(this->*word_processor)(&page_res_it))
return;
}
}
}
void apply_box_training(BLOCK_LIST *block_list) {
BLOCK_IT block_it(block_list);
ROW_IT row_it;
ROW *row;
WERD_IT word_it;
WERD *word;
WERD *bln_word;
WERD copy_outword; // copy to denorm
PBLOB_IT blob_it;
DENORM denorm;
INT16 count = 0;
char ch[2];
ch[1] = '\0';
tprintf ("Generating training data\n");
for (block_it.mark_cycle_pt ();
!block_it.cycled_list (); block_it.forward ()) {
row_it.set_to_list (block_it.data ()->row_list ());
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
word_it.set_to_list (row->word_list ());
for (word_it.mark_cycle_pt ();
!word_it.cycled_list (); word_it.forward ()) {
word = word_it.data ();
if ((strlen (word->text ()) == 1) &&
(word->gblob_list ()->length () == 1)) {
/* Here is a word with a single char label and a single blob so train on it */
bln_word =
make_bln_copy (word, row, row->x_height (), &denorm);
blob_it.set_to_list (bln_word->blob_list ());
ch[0] = *word->text ();
tess_training_tester (blob_it.data (),
//single blob
&denorm, TRUE, //correct
ch, //correct ASCII char
1, //ASCII length
NULL);
copy_outword = *(bln_word);
copy_outword.baseline_denormalise (&denorm);
blob_it.set_to_list (copy_outword.blob_list ());
ch[0] = *word->text ();
delete bln_word;
count++;
}
}
}
}
tprintf ("Generated training data for %d blobs\n", count);
}
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;
}
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);
}
}
}
/**
* Returns the baseline of the current object at the given level.
* The baseline is the line that passes through (x1, y1) and (x2, y2).
* WARNING: with vertical text, baselines may be vertical!
*/
bool PageIterator::Baseline(PageIteratorLevel level,
int* x1, int* y1, int* x2, int* y2) const {
if (it_->word() == NULL) return false; // Already at the end!
ROW* row = it_->row()->row;
WERD* word = it_->word()->word;
TBOX box = (level == RIL_WORD || level == RIL_SYMBOL)
? word->bounding_box()
: row->bounding_box();
int left = box.left();
ICOORD startpt(left, static_cast<inT16>(row->base_line(left) + 0.5));
int right = box.right();
ICOORD endpt(right, static_cast<inT16>(row->base_line(right) + 0.5));
// Rotate to image coordinates and convert to global image coords.
startpt.rotate(it_->block()->block->re_rotation());
endpt.rotate(it_->block()->block->re_rotation());
*x1 = startpt.x() / scale_ + rect_left_;
*y1 = (rect_height_ - startpt.y()) / scale_ + rect_top_;
*x2 = endpt.x() / scale_ + rect_left_;
*y2 = (rect_height_ - endpt.y()) / scale_ + rect_top_;
return true;
}
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();
}
}
/**
* word_set_display() Word processor
*
* Display word according to current display mode settings
*/
BOOL8 Tesseract::word_set_display(BLOCK* block, ROW* row, WERD_RES* word_res) {
WERD* word = word_res->word;
word->set_display_flag(DF_BOX, word_display_mode.bit(DF_BOX));
word->set_display_flag(DF_TEXT, word_display_mode.bit(DF_TEXT));
word->set_display_flag(DF_POLYGONAL, word_display_mode.bit(DF_POLYGONAL));
word->set_display_flag(DF_EDGE_STEP, word_display_mode.bit(DF_EDGE_STEP));
word->set_display_flag(DF_BN_POLYGONAL,
word_display_mode.bit(DF_BN_POLYGONAL));
word->set_display_flag(DF_BLAMER, word_display_mode.bit(DF_BLAMER));
return word_display(block, row, word_res);
}
/**
* word_set_display() Word processor
*
* Display word according to current display mode settings
*/
BOOL8 Tesseract::word_set_display(PAGE_RES_IT* pr_it) {
WERD* word = pr_it->word()->word;
word->set_display_flag(DF_BOX, word_display_mode.bit(DF_BOX));
word->set_display_flag(DF_TEXT, word_display_mode.bit(DF_TEXT));
word->set_display_flag(DF_POLYGONAL, word_display_mode.bit(DF_POLYGONAL));
word->set_display_flag(DF_EDGE_STEP, word_display_mode.bit(DF_EDGE_STEP));
word->set_display_flag(DF_BN_POLYGONAL,
word_display_mode.bit(DF_BN_POLYGONAL));
word->set_display_flag(DF_BLAMER, word_display_mode.bit(DF_BLAMER));
return word_display(pr_it);
}
void ROW::recalc_bounding_box() { //recalculate BB
WERD *word; //current word
WERD_IT it = &words; //words of ROW
inT16 left; //of word
inT16 prev_left; //old left
if (!it.empty ()) {
word = it.data ();
prev_left = word->bounding_box ().left ();
it.forward ();
while (!it.at_first ()) {
word = it.data ();
left = word->bounding_box ().left ();
if (left < prev_left) {
it.move_to_first ();
//words in BB order
it.sort (word_comparator);
break;
}
prev_left = left;
it.forward ();
}
}
for (it.mark_cycle_pt (); !it.cycled_list (); it.forward ()) {
word = it.data ();
if (it.at_first ())
word->set_flag (W_BOL, TRUE);
else
//not start of line
word->set_flag (W_BOL, FALSE);
if (it.at_last ())
word->set_flag (W_EOL, TRUE);
else
//not end of line
word->set_flag (W_EOL, FALSE);
//extend BB as reqd
bound_box += word->bounding_box ();
}
}
/// Resegments the words by running the classifier in an attempt to find the
/// correct segmentation that produces the required string.
void Tesseract::ReSegmentByClassification(PAGE_RES* page_res) {
PAGE_RES_IT pr_it(page_res);
WERD_RES* word_res;
for (; (word_res = pr_it.word()) != NULL; pr_it.forward()) {
WERD* word = word_res->word;
if (word->text() == NULL || word->text()[0] == '\0')
continue; // Ignore words that have no text.
// Convert the correct text to a vector of UNICHAR_ID
GenericVector<UNICHAR_ID> target_text;
if (!ConvertStringToUnichars(word->text(), &target_text)) {
tprintf("APPLY_BOX: FAILURE: can't find class_id for '%s'\n",
word->text());
pr_it.DeleteCurrentWord();
continue;
}
if (!FindSegmentation(target_text, word_res)) {
tprintf("APPLY_BOX: FAILURE: can't find segmentation for '%s'\n",
word->text());
pr_it.DeleteCurrentWord();
continue;
}
}
}
/**********************************************************************
* fixed_pitch_words
*
* Make a ROW from a fixed pitch TO_ROW.
**********************************************************************/
ROW *fixed_pitch_words( //find lines
TO_ROW *row, //row to do
FCOORD rotation //for drawing
) {
bool bol; //start of line
uint8_t blanks; //in front of word
uint8_t new_blanks; //blanks in empty cell
int16_t chop_coord; //chop boundary
int16_t prev_chop_coord; //start of cell
int16_t rep_left; //left edge of rep word
ROW *real_row; //output row
C_OUTLINE_LIST left_coutlines;
C_OUTLINE_LIST right_coutlines;
C_BLOB_LIST cblobs;
C_BLOB_IT cblob_it = &cblobs;
WERD_LIST words;
WERD_IT word_it = &words; //new words
//repeated blobs
WERD_IT rep_it = &row->rep_words;
WERD *word; //new word
int32_t xstarts[2]; //row ends
int32_t prev_x; //end of prev blob
//iterator
BLOBNBOX_IT box_it = row->blob_list ();
//boundaries
ICOORDELT_IT cell_it = &row->char_cells;
#ifndef GRAPHICS_DISABLED
if (textord_show_page_cuts && to_win != nullptr) {
plot_row_cells (to_win, ScrollView::RED, row, 0, &row->char_cells);
}
#endif
prev_x = -INT16_MAX;
bol = true;
blanks = 0;
if (rep_it.empty ())
rep_left = INT16_MAX;
else
rep_left = rep_it.data ()->bounding_box ().left ();
if (box_it.empty ())
return nullptr; //empty row
xstarts[0] = box_it.data ()->bounding_box ().left ();
if (rep_left < xstarts[0]) {
xstarts[0] = rep_left;
}
if (cell_it.empty () || row->char_cells.singleton ()) {
tprintf ("Row without enough char cells!\n");
tprintf ("Leftmost blob is at (%d,%d)\n",
box_it.data ()->bounding_box ().left (),
box_it.data ()->bounding_box ().bottom ());
return nullptr;
}
ASSERT_HOST (!cell_it.empty () && !row->char_cells.singleton ());
prev_chop_coord = cell_it.data ()->x ();
word = nullptr;
while (rep_left < cell_it.data ()->x ()) {
word = add_repeated_word (&rep_it, rep_left, prev_chop_coord,
blanks, row->fixed_pitch, &word_it);
}
cell_it.mark_cycle_pt ();
if (prev_chop_coord >= cell_it.data ()->x ())
cell_it.forward ();
for (; !cell_it.cycled_list (); cell_it.forward ()) {
chop_coord = cell_it.data ()->x ();
while (!box_it.empty ()
&& box_it.data ()->bounding_box ().left () <= chop_coord) {
if (box_it.data ()->bounding_box ().right () > prev_x)
prev_x = box_it.data ()->bounding_box ().right ();
split_to_blob (box_it.extract (), chop_coord,
textord_fp_chop_error + 0.5f,
&left_coutlines,
&right_coutlines);
box_it.forward ();
while (!box_it.empty() && box_it.data()->cblob() == nullptr) {
delete box_it.extract();
box_it.forward();
}
}
if (!right_coutlines.empty() && left_coutlines.empty())
split_to_blob (nullptr, chop_coord,
textord_fp_chop_error + 0.5f,
&left_coutlines,
&right_coutlines);
if (!left_coutlines.empty()) {
cblob_it.add_after_then_move(new C_BLOB(&left_coutlines));
} else {
if (rep_left < chop_coord) {
if (rep_left > prev_chop_coord)
new_blanks = (uint8_t) floor ((rep_left - prev_chop_coord)
/ row->fixed_pitch + 0.5);
else
new_blanks = 0;
}
else {
if (chop_coord > prev_chop_coord)
new_blanks = (uint8_t) floor ((chop_coord - prev_chop_coord)
/ row->fixed_pitch + 0.5);
else
new_blanks = 0;
}
if (!cblob_it.empty()) {
if (blanks < 1 && word != nullptr && !word->flag (W_REP_CHAR))
blanks = 1;
word = new WERD (&cblobs, blanks, nullptr);
cblob_it.set_to_list (&cblobs);
word->set_flag (W_DONT_CHOP, TRUE);
word_it.add_after_then_move (word);
if (bol) {
word->set_flag (W_BOL, TRUE);
bol = false;
}
blanks = new_blanks;
}
else
blanks += new_blanks;
while (rep_left < chop_coord) {
word = add_repeated_word (&rep_it, rep_left, prev_chop_coord,
blanks, row->fixed_pitch, &word_it);
}
}
if (prev_chop_coord < chop_coord)
prev_chop_coord = chop_coord;
}
if (!cblob_it.empty()) {
word = new WERD(&cblobs, blanks, nullptr);
word->set_flag (W_DONT_CHOP, TRUE);
word_it.add_after_then_move (word);
if (bol)
word->set_flag (W_BOL, TRUE);
}
ASSERT_HOST (word != nullptr);
while (!rep_it.empty ()) {
add_repeated_word (&rep_it, rep_left, prev_chop_coord,
blanks, row->fixed_pitch, &word_it);
}
//at end of line
word_it.data ()->set_flag (W_EOL, TRUE);
if (prev_chop_coord > prev_x)
prev_x = prev_chop_coord;
xstarts[1] = prev_x + 1;
real_row = new ROW (row, (int16_t) row->kern_size, (int16_t) row->space_size);
word_it.set_to_list (real_row->word_list ());
//put words in row
word_it.add_list_after (&words);
real_row->recalc_bounding_box ();
return real_row;
}
/// Consume all source blobs that strongly overlap the given box,
/// putting them into a new word, with the correct_text label.
/// Fights over which box owns which blobs are settled by
/// applying the blobs to box or next_box with the least non-overlap.
/// @return false if the box was in error, which can only be caused by
/// failing to find an overlapping blob for a box.
bool Tesseract::ResegmentWordBox(BLOCK_LIST *block_list,
const TBOX& box, const TBOX& next_box,
const char* correct_text) {
if (applybox_debug > 1) {
tprintf("\nAPPLY_BOX: in ResegmentWordBox() for %s\n", correct_text);
}
WERD* new_word = NULL;
BLOCK_IT b_it(block_list);
for (b_it.mark_cycle_pt(); !b_it.cycled_list(); b_it.forward()) {
BLOCK* block = b_it.data();
if (!box.major_overlap(block->bounding_box()))
continue;
ROW_IT r_it(block->row_list());
for (r_it.mark_cycle_pt(); !r_it.cycled_list(); r_it.forward()) {
ROW* row = r_it.data();
if (!box.major_overlap(row->bounding_box()))
continue;
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();
if (applybox_debug > 2) {
tprintf("Checking word:");
word->bounding_box().print();
}
if (word->text() != NULL && word->text()[0] != '\0')
continue; // Ignore words that are already done.
if (!box.major_overlap(word->bounding_box()))
continue;
C_BLOB_IT blob_it(word->cblob_list());
for (blob_it.mark_cycle_pt(); !blob_it.cycled_list();
blob_it.forward()) {
C_BLOB* blob = blob_it.data();
TBOX blob_box = blob->bounding_box();
if (!blob_box.major_overlap(box))
continue;
double current_box_miss_metric = BoxMissMetric(blob_box, box);
double next_box_miss_metric = BoxMissMetric(blob_box, next_box);
if (applybox_debug > 2) {
tprintf("Checking blob:");
blob_box.print();
tprintf("Current miss metric = %g, next = %g\n",
current_box_miss_metric, next_box_miss_metric);
}
if (current_box_miss_metric > next_box_miss_metric)
continue; // Blob is a better match for next box.
if (applybox_debug > 2) {
tprintf("Blob match: blob:");
blob_box.print();
tprintf("Matches box:");
box.print();
tprintf("With next box:");
next_box.print();
}
if (new_word == NULL) {
// Make a new word with a single blob.
new_word = word->shallow_copy();
new_word->set_text(correct_text);
w_it.add_to_end(new_word);
}
C_BLOB_IT new_blob_it(new_word->cblob_list());
new_blob_it.add_to_end(blob_it.extract());
}
}
}
}
if (new_word == NULL && applybox_debug > 0) tprintf("FAIL!\n");
return new_word != NULL;
}
/**
* word_display() Word Processor
*
* Display a word according to its display modes
*/
BOOL8 Tesseract::word_display(BLOCK* block, ROW* row, WERD_RES* word_res) {
WERD* word = word_res->word;
TBOX word_bb; // word bounding box
int word_height; // ht of word BB
BOOL8 displayed_something = FALSE;
float shift; // from bot left
C_BLOB_IT c_it; // cblob iterator
if (color_mode != CM_RAINBOW && word_res->box_word != NULL) {
BoxWord* box_word = word_res->box_word;
int length = box_word->length();
if (word_res->fontinfo == NULL) return false;
const FontInfo& font_info = *word_res->fontinfo;
for (int i = 0; i < length; ++i) {
ScrollView::Color color = ScrollView::GREEN;
switch (color_mode) {
case CM_SUBSCRIPT:
if (box_word->BlobPosition(i) == SP_SUBSCRIPT)
color = ScrollView::RED;
break;
case CM_SUPERSCRIPT:
if (box_word->BlobPosition(i) == SP_SUPERSCRIPT)
color = ScrollView::RED;
break;
case CM_ITALIC:
if (font_info.is_italic())
color = ScrollView::RED;
break;
case CM_BOLD:
if (font_info.is_bold())
color = ScrollView::RED;
break;
case CM_FIXEDPITCH:
if (font_info.is_fixed_pitch())
color = ScrollView::RED;
break;
case CM_SERIF:
if (font_info.is_serif())
color = ScrollView::RED;
break;
case CM_SMALLCAPS:
if (word_res->small_caps)
color = ScrollView::RED;
break;
case CM_DROPCAPS:
if (box_word->BlobPosition(i) == SP_DROPCAP)
color = ScrollView::RED;
break;
// TODO(rays) underline is currently completely unsupported.
case CM_UNDERLINE:
default:
break;
}
image_win->Pen(color);
TBOX box = box_word->BlobBox(i);
image_win->Rectangle(box.left(), box.bottom(), box.right(), box.top());
}
return true;
}
/*
Note the double coercions of(COLOUR)((inT32)editor_image_word_bb_color)
etc. are to keep the compiler happy.
*/
// display bounding box
if (word->display_flag(DF_BOX)) {
word->bounding_box().plot(image_win,
(ScrollView::Color)((inT32)
editor_image_word_bb_color),
(ScrollView::Color)((inT32)
editor_image_word_bb_color));
ScrollView::Color c = (ScrollView::Color)
((inT32) editor_image_blob_bb_color);
image_win->Pen(c);
c_it.set_to_list(word->cblob_list());
for (c_it.mark_cycle_pt(); !c_it.cycled_list(); c_it.forward())
c_it.data()->bounding_box().plot(image_win);
displayed_something = TRUE;
}
// display edge steps
if (word->display_flag(DF_EDGE_STEP)) { // edgesteps available
word->plot(image_win); // rainbow colors
displayed_something = TRUE;
}
// display poly approx
if (word->display_flag(DF_POLYGONAL)) {
// need to convert
TWERD* tword = TWERD::PolygonalCopy(word);
tword->plot(image_win);
delete tword;
displayed_something = TRUE;
}
// Display correct text and blamer information.
STRING text;
STRING blame;
if (word->display_flag(DF_TEXT) && word->text() != NULL) {
text = word->text();
}
if (word->display_flag(DF_BLAMER) &&
!(word_res->blamer_bundle != NULL &&
word_res->blamer_bundle->incorrect_result_reason == IRR_CORRECT)) {
text = "";
const BlamerBundle *blamer_bundle = word_res->blamer_bundle;
if (blamer_bundle == NULL) {
text += "NULL";
} else {
for (int i = 0; i < blamer_bundle->truth_text.length(); ++i) {
text += blamer_bundle->truth_text[i];
}
}
text += " -> ";
STRING best_choice_str;
if (word_res->best_choice == NULL) {
best_choice_str = "NULL";
} else {
word_res->best_choice->string_and_lengths(&best_choice_str, NULL);
}
text += best_choice_str;
IncorrectResultReason reason = (blamer_bundle == NULL) ?
IRR_PAGE_LAYOUT : blamer_bundle->incorrect_result_reason;
ASSERT_HOST(reason < IRR_NUM_REASONS)
blame += " [";
blame += BlamerBundle::IncorrectReasonName(reason);
blame += "]";
}
if (text.length() > 0) {
word_bb = word->bounding_box();
image_win->Pen(ScrollView::RED);
word_height = word_bb.height();
int text_height = 0.50 * word_height;
if (text_height > 20) text_height = 20;
image_win->TextAttributes("Arial", text_height, false, false, false);
shift = (word_height < word_bb.width()) ? 0.25 * word_height : 0.0f;
image_win->Text(word_bb.left() + shift,
word_bb.bottom() + 0.25 * word_height, text.string());
if (blame.length() > 0) {
image_win->Text(word_bb.left() + shift,
word_bb.bottom() + 0.25 * word_height - text_height,
blame.string());
}
displayed_something = TRUE;
}
if (!displayed_something) // display BBox anyway
word->bounding_box().plot(image_win,
(ScrollView::Color)((inT32) editor_image_word_bb_color),
(ScrollView::Color)((inT32)
editor_image_word_bb_color));
return TRUE;
}
void apply_box_testing(BLOCK_LIST *block_list) {
BLOCK_IT block_it(block_list);
ROW_IT row_it;
ROW *row;
INT16 row_count = 0;
WERD_IT word_it;
WERD *word;
WERD *bln_word;
INT16 word_count = 0;
PBLOB_IT blob_it;
DENORM denorm;
INT16 count = 0;
char ch[2];
WERD *outword; //bln best choice
//segmentation
WERD_CHOICE *best_choice; //tess output
WERD_CHOICE *raw_choice; //top choice permuter
//detailed results
BLOB_CHOICE_LIST_CLIST blob_choices;
INT16 char_count = 0;
INT16 correct_count = 0;
INT16 err_count = 0;
INT16 rej_count = 0;
#ifndef SECURE_NAMES
WERDSTATS wordstats; //As from newdiff
#endif
char tess_rej_str[3];
char tess_long_str[3];
ch[1] = '\0';
strcpy (tess_rej_str, "|A");
strcpy (tess_long_str, "|B");
for (block_it.mark_cycle_pt ();
!block_it.cycled_list (); block_it.forward ()) {
row_it.set_to_list (block_it.data ()->row_list ());
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
row_count++;
word_count = 0;
word_it.set_to_list (row->word_list ());
for (word_it.mark_cycle_pt ();
!word_it.cycled_list (); word_it.forward ()) {
word = word_it.data ();
word_count++;
if ((strlen (word->text ()) == 1) &&
!STRING (applybox_test_exclusions).contains (*word->text ())
&& (word->gblob_list ()->length () == 1)) {
/* Here is a word with a single char label and a single blob so test it */
bln_word =
make_bln_copy (word, row, row->x_height (), &denorm);
blob_it.set_to_list (bln_word->blob_list ());
ch[0] = *word->text ();
char_count++;
best_choice = tess_segment_pass1 (bln_word,
&denorm,
tess_default_matcher,
raw_choice,
&blob_choices, outword);
/*
Test for TESS screw up on word. Recog_word has already ensured that the
choice list, outword blob lists and best_choice string are the same
length. A TESS screw up is indicated by a blank filled or 0 length string.
*/
if ((best_choice->string ().length () == 0) ||
(strspn (best_choice->string ().string (), " ") ==
best_choice->string ().length ())) {
rej_count++;
tprintf ("%d:%d: \"%s\" -> TESS FAILED\n",
row_count, word_count, ch);
#ifndef SECURE_NAMES
wordstats.word (tess_rej_str, 2, ch, 1);
#endif
}
else {
if ((best_choice->string ().length () !=
outword->blob_list ()->length ()) ||
(best_choice->string ().length () !=
blob_choices.length ())) {
tprintf
("ASSERT FAIL String:\"%s\"; Strlen=%d; #Blobs=%d; #Choices=%d\n",
best_choice->string ().string (),
best_choice->string ().length (),
outword->blob_list ()->length (),
blob_choices.length ());
}
ASSERT_HOST (best_choice->string ().length () ==
outword->blob_list ()->length ());
ASSERT_HOST (best_choice->string ().length () ==
blob_choices.length ());
fix_quotes ((char *) best_choice->string ().string (),
//turn to double
outword, &blob_choices);
if (strcmp (best_choice->string ().string (), ch) != 0) {
err_count++;
tprintf ("%d:%d: \"%s\" -> \"%s\"\n",
row_count, word_count, ch,
best_choice->string ().string ());
}
else
correct_count++;
#ifndef SECURE_NAMES
if (best_choice->string ().length () > 2)
wordstats.word (tess_long_str, 2, ch, 1);
else
wordstats.word ((char *) best_choice->string ().
string (),
best_choice->string ().length (), ch,
1);
#endif
}
delete bln_word;
delete outword;
delete best_choice;
delete raw_choice;
blob_choices.deep_clear ();
count++;
}
}
}
}
#ifndef SECURE_NAMES
wordstats.print (1, 100.0);
wordstats.conf_matrix ();
tprintf ("Tested %d chars: %d correct; %d rejected by tess; %d errs\n",
char_count, correct_count, rej_count, err_count);
#endif
}
/*************************************************************************
* tidy_up()
* - report >1 block
* - sort the words in each row.
* - report any rows with no labelled words.
* - report any remaining unlabelled words
* - report total labelled words
*
*************************************************************************/
void tidy_up( //
BLOCK_LIST *block_list, //real blocks
INT16 &ok_char_count,
INT16 &ok_row_count,
INT16 &unlabelled_words,
INT16 *tgt_char_counts,
INT16 &rebalance_count,
char &min_char,
INT16 &min_samples,
INT16 &final_labelled_blob_count) {
BLOCK_IT block_it(block_list);
ROW_IT row_it;
ROW *row;
WERD_IT word_it;
WERD *word;
WERD *duplicate_word;
INT16 block_idx = 0;
INT16 row_idx;
INT16 all_row_idx = 0;
BOOL8 row_ok;
BOOL8 rebalance_needed = FALSE;
//No. of unique labelled samples
INT16 labelled_char_counts[128];
INT16 i;
char ch;
char prev_ch = '\0';
BOOL8 at_dupe_of_prev_word;
ROW *prev_row = NULL;
INT16 left;
INT16 prev_left = -1;
for (i = 0; i < 128; i++)
labelled_char_counts[i] = 0;
ok_char_count = 0;
ok_row_count = 0;
unlabelled_words = 0;
if ((applybox_debug > 4) && (block_it.length () != 1))
tprintf ("APPLY_BOXES: More than one block??\n");
for (block_it.mark_cycle_pt ();
!block_it.cycled_list (); block_it.forward ()) {
block_idx++;
row_idx = 0;
row_ok = FALSE;
row_it.set_to_list (block_it.data ()->row_list ());
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row_idx++;
all_row_idx++;
row = row_it.data ();
word_it.set_to_list (row->word_list ());
word_it.sort (word_comparator);
for (word_it.mark_cycle_pt ();
!word_it.cycled_list (); word_it.forward ()) {
word = word_it.data ();
if (strlen (word->text ()) == 0) {
unlabelled_words++;
if (applybox_debug > 4) {
tprintf
("APPLY_BOXES: Unlabelled word blk:%d row:%d allrows:%d\n",
block_idx, row_idx, all_row_idx);
}
}
else {
if (word->gblob_list ()->length () != 1)
tprintf
("APPLY_BOXES: FATALITY - MULTIBLOB Labelled word blk:%d row:%d allrows:%d\n",
block_idx, row_idx, all_row_idx);
ok_char_count++;
labelled_char_counts[*word->text ()]++;
row_ok = TRUE;
}
}
if ((applybox_debug > 4) && (!row_ok)) {
tprintf
("APPLY_BOXES: Row with no labelled words blk:%d row:%d allrows:%d\n",
block_idx, row_idx, all_row_idx);
}
else
ok_row_count++;
}
}
min_samples = 9999;
for (i = 0; i < 128; i++) {
if (tgt_char_counts[i] > labelled_char_counts[i]) {
if (labelled_char_counts[i] <= 1) {
tprintf
("APPLY_BOXES: FATALITY - %d labelled samples of \"%c\" - target is %d\n",
labelled_char_counts[i], (char) i, tgt_char_counts[i]);
}
else {
rebalance_needed = TRUE;
if (applybox_debug > 0)
tprintf
("APPLY_BOXES: REBALANCE REQD \"%c\" - target of %d from %d labelled samples\n",
(char) i, tgt_char_counts[i], labelled_char_counts[i]);
}
}
if ((min_samples > labelled_char_counts[i]) && (tgt_char_counts[i] > 0)) {
min_samples = labelled_char_counts[i];
min_char = (char) i;
}
}
while (applybox_rebalance && rebalance_needed) {
block_it.set_to_list (block_list);
for (block_it.mark_cycle_pt ();
!block_it.cycled_list (); block_it.forward ()) {
row_it.set_to_list (block_it.data ()->row_list ());
for (row_it.mark_cycle_pt ();
!row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
word_it.set_to_list (row->word_list ());
for (word_it.mark_cycle_pt ();
!word_it.cycled_list (); word_it.forward ()) {
word = word_it.data ();
left = word->bounding_box ().left ();
ch = *word->text ();
at_dupe_of_prev_word = ((row == prev_row) &&
(left = prev_left) &&
(ch == prev_ch));
if ((ch != '\0') &&
(labelled_char_counts[ch] > 1) &&
(tgt_char_counts[ch] > labelled_char_counts[ch]) &&
(!at_dupe_of_prev_word)) {
/* Duplicate the word to rebalance the labelled samples */
if (applybox_debug > 9) {
tprintf ("Duping \"%c\" from ", ch);
word->bounding_box ().print ();
}
duplicate_word = new WERD;
*duplicate_word = *word;
word_it.add_after_then_move (duplicate_word);
rebalance_count++;
labelled_char_counts[ch]++;
}
prev_row = row;
prev_left = left;
prev_ch = ch;
}
}
}
rebalance_needed = FALSE;
for (i = 0; i < 128; i++) {
if ((tgt_char_counts[i] > labelled_char_counts[i]) &&
(labelled_char_counts[i] > 1)) {
rebalance_needed = TRUE;
break;
}
}
}
/* Now final check - count labelled blobs */
final_labelled_blob_count = 0;
block_it.set_to_list (block_list);
for (block_it.mark_cycle_pt ();
!block_it.cycled_list (); block_it.forward ()) {
row_it.set_to_list (block_it.data ()->row_list ());
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
word_it.set_to_list (row->word_list ());
word_it.sort (word_comparator);
for (word_it.mark_cycle_pt ();
!word_it.cycled_list (); word_it.forward ()) {
word = word_it.data ();
if ((strlen (word->text ()) == 1) &&
(word->gblob_list ()->length () == 1))
final_labelled_blob_count++;
}
}
}
}
INT16 resegment_box( //
ROW *row,
BOX box,
char *ch,
INT16 block_id,
INT16 row_id,
INT16 boxfile_lineno,
INT16 boxfile_charno) {
WERD_IT word_it;
WERD *word;
WERD *new_word = NULL;
BOOL8 polyg = false;
PBLOB_IT blob_it;
PBLOB_IT new_blob_it;
PBLOB *blob;
PBLOB *new_blob;
OUTLINE_IT outline_it;
OUTLINE_LIST dummy; // Just to initialize new_outline_it.
OUTLINE_IT new_outline_it = &dummy;
OUTLINE *outline;
BOX new_word_box;
float word_x_centre;
float baseline;
INT16 error_count = 0; //number of chars lost
word_it.set_to_list (row->word_list ());
for (word_it.mark_cycle_pt (); !word_it.cycled_list (); word_it.forward ()) {
word = word_it.data ();
polyg = word->flag (W_POLYGON);
if (word->bounding_box ().overlap (box)) {
blob_it.set_to_list (word->gblob_list ());
for (blob_it.mark_cycle_pt ();
!blob_it.cycled_list (); blob_it.forward ()) {
blob = blob_it.data ();
if (gblob_bounding_box (blob, polyg).overlap (box)) {
outline_it.set_to_list (gblob_out_list (blob, polyg));
for (outline_it.mark_cycle_pt ();
!outline_it.cycled_list (); outline_it.forward ()) {
outline = outline_it.data ();
if (goutline_bounding_box (outline, polyg).
major_overlap (box)) {
if (strlen (word->text ()) > 0) {
if (error_count == 0) {
error_count = 1;
if (applybox_debug > 4)
report_failed_box (boxfile_lineno,
boxfile_charno,
box, ch,
"FAILURE! box overlaps blob in labelled word");
}
if (applybox_debug > 4)
tprintf
("APPLY_BOXES: ALSO ignoring corrupted char blk:%d row:%d \"%s\"\n",
block_id, row_id,
word_it.data ()->text ());
word_it.data ()->set_text ("");
//UN label it
error_count++;
}
if (error_count == 0) {
if (new_word == NULL) {
/* Make a new word with a single blob */
new_word = word->shallow_copy ();
new_word->set_text (ch);
if (polyg)
new_blob = new PBLOB;
else
new_blob = (PBLOB *) new C_BLOB;
new_blob_it.set_to_list (new_word->
gblob_list ());
new_blob_it.add_to_end (new_blob);
new_outline_it.
set_to_list (gblob_out_list
(new_blob, polyg));
}
new_outline_it.add_to_end (outline_it.
extract ());
//move blob
}
}
}
//no outlines in blob
if (outline_it.empty ())
//so delete blob
delete blob_it.extract ();
}
}
if (blob_it.empty ()) //no blobs in word
//so delete word
delete word_it.extract ();
}
}
if (error_count > 0)
return error_count;
if (new_word != NULL) {
gblob_sort_list (new_word->gblob_list (), polyg);
word_it.add_to_end (new_word);
new_word_box = new_word->bounding_box ();
word_x_centre = (new_word_box.left () + new_word_box.right ()) / 2.0f;
baseline = row->base_line (word_x_centre);
if (STRING (chs_caps_ht).contains (ch[0]) &&
(new_word_box.top () <
baseline + (1 + applybox_error_band) * row->x_height ())) {
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! caps-ht char didn't ascend");
new_word->set_text ("");
return 1;
}
if (STRING (chs_odd_top).contains (ch[0]) &&
(new_word_box.top () <
baseline + (1 - applybox_error_band) * row->x_height ())) {
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! Odd top char below xht");
new_word->set_text ("");
return 1;
}
if (STRING (chs_x_ht).contains (ch[0]) &&
((new_word_box.top () >
baseline + (1 + applybox_error_band) * row->x_height ()) ||
(new_word_box.top () <
baseline + (1 - applybox_error_band) * row->x_height ()))) {
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! x-ht char didn't have top near xht");
new_word->set_text ("");
return 1;
}
if (STRING (chs_non_ambig_bl).contains (ch[0]) &&
((new_word_box.bottom () <
baseline - applybox_error_band * row->x_height ()) ||
(new_word_box.bottom () >
baseline + applybox_error_band * row->x_height ()))) {
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! non ambig BL char didnt have bottom near baseline");
new_word->set_text ("");
return 1;
}
if (STRING (chs_odd_bot).contains (ch[0]) &&
(new_word_box.bottom () >
baseline + applybox_error_band * row->x_height ())) {
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! Odd bottom char above baseline");
new_word->set_text ("");
return 1;
}
if (STRING (chs_desc).contains (ch[0]) &&
(new_word_box.bottom () >
baseline - applybox_error_band * row->x_height ())) {
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! Descender doesn't descend");
new_word->set_text ("");
return 1;
}
return 0;
}
else {
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! Couldn't find any blobs");
return 1;
}
}
ROW *find_row_of_box( //
BLOCK_LIST *block_list, //real blocks
BOX box, //from boxfile
INT16 &block_id,
INT16 &row_id_to_process) {
BLOCK_IT block_it(block_list);
BLOCK *block;
ROW_IT row_it;
ROW *row;
ROW *row_to_process = NULL;
INT16 row_id;
WERD_IT word_it;
WERD *word;
BOOL8 polyg;
PBLOB_IT blob_it;
PBLOB *blob;
OUTLINE_IT outline_it;
OUTLINE *outline;
/*
Find row to process - error if box REALLY overlaps more than one row. (I.e
it overlaps blobs in the row - not just overlaps the bounding box of the
whole row.)
*/
block_id = 0;
for (block_it.mark_cycle_pt ();
!block_it.cycled_list (); block_it.forward ()) {
block_id++;
row_id = 0;
block = block_it.data ();
if (block->bounding_box ().overlap (box)) {
row_it.set_to_list (block->row_list ());
for (row_it.mark_cycle_pt ();
!row_it.cycled_list (); row_it.forward ()) {
row_id++;
row = row_it.data ();
if (row->bounding_box ().overlap (box)) {
word_it.set_to_list (row->word_list ());
for (word_it.mark_cycle_pt ();
!word_it.cycled_list (); word_it.forward ()) {
word = word_it.data ();
polyg = word->flag (W_POLYGON);
if (word->bounding_box ().overlap (box)) {
blob_it.set_to_list (word->gblob_list ());
for (blob_it.mark_cycle_pt ();
!blob_it.cycled_list (); blob_it.forward ()) {
blob = blob_it.data ();
if (gblob_bounding_box (blob, polyg).
overlap (box)) {
outline_it.
set_to_list (gblob_out_list
(blob, polyg));
for (outline_it.mark_cycle_pt ();
!outline_it.cycled_list ();
outline_it.forward ()) {
outline = outline_it.data ();
if (goutline_bounding_box
(outline, polyg).major_overlap (box)) {
if ((row_to_process == NULL) ||
(row_to_process == row)) {
row_to_process = row;
row_id_to_process = row_id;
}
else
/* RETURN ERROR Box overlaps blobs in more than one row */
return NULL;
}
}
}
}
}
}
}
}
}
}
return row_to_process;
}
bool Textord::clean_noise_from_row( //remove empties
ROW* row //row to clean
) {
bool testing_on;
TBOX blob_box; //bounding box
C_BLOB *blob; //current blob
C_OUTLINE *outline; //current outline
WERD *word; //current word
int32_t blob_size; //biggest size
int32_t trans_count = 0; //no of transitions
int32_t trans_threshold; //noise tolerance
int32_t dot_count; //small objects
int32_t norm_count; //normal objects
int32_t super_norm_count; //real char-like
//words of row
WERD_IT word_it = row->word_list ();
C_BLOB_IT blob_it; //blob iterator
C_OUTLINE_IT out_it; //outline iterator
testing_on = textord_test_y > row->base_line (textord_test_x)
&& textord_show_blobs
&& textord_test_y < row->base_line (textord_test_x) + row->x_height ();
dot_count = 0;
norm_count = 0;
super_norm_count = 0;
for (word_it.mark_cycle_pt (); !word_it.cycled_list (); word_it.forward ()) {
word = word_it.data (); //current word
//blobs in word
blob_it.set_to_list (word->cblob_list ());
for (blob_it.mark_cycle_pt (); !blob_it.cycled_list ();
blob_it.forward ()) {
blob = blob_it.data ();
if (!word->flag (W_DONT_CHOP)) {
//get outlines
out_it.set_to_list (blob->out_list ());
for (out_it.mark_cycle_pt (); !out_it.cycled_list ();
out_it.forward ()) {
outline = out_it.data ();
blob_box = outline->bounding_box ();
blob_size =
blob_box.width () >
blob_box.height ()? blob_box.width () : blob_box.
height();
if (blob_size < textord_noise_sizelimit * row->x_height ())
dot_count++; //count smal outlines
if (!outline->child ()->empty ()
&& blob_box.height () <
(1 + textord_noise_syfract) * row->x_height ()
&& blob_box.height () >
(1 - textord_noise_syfract) * row->x_height ()
&& blob_box.width () <
(1 + textord_noise_sxfract) * row->x_height ()
&& blob_box.width () >
(1 - textord_noise_sxfract) * row->x_height ())
super_norm_count++; //count smal outlines
}
}
else
super_norm_count++;
blob_box = blob->bounding_box ();
blob_size =
blob_box.width () >
blob_box.height ()? blob_box.width () : blob_box.height ();
if (blob_size >= textord_noise_sizelimit * row->x_height ()
&& blob_size < row->x_height () * 2) {
trans_threshold = blob_size / textord_noise_sizefraction;
trans_count = blob->count_transitions (trans_threshold);
if (trans_count < textord_noise_translimit)
norm_count++;
}
else if (blob_box.height () > row->x_height () * 2
&& (!word_it.at_first () || !blob_it.at_first ()))
dot_count += 2;
if (testing_on) {
tprintf
("Blob at (%d,%d) -> (%d,%d), ols=%d, tc=%d, bldiff=%g\n",
blob_box.left (), blob_box.bottom (), blob_box.right (),
blob_box.top (), blob->out_list ()->length (), trans_count,
blob_box.bottom () - row->base_line (blob_box.left ()));
}
}
}
if (textord_noise_debug) {
tprintf ("Row ending at (%d,%g):",
blob_box.right (), row->base_line (blob_box.right ()));
tprintf (" R=%g, dc=%d, nc=%d, %s\n",
norm_count > 0 ? (float) dot_count / norm_count : 9999,
dot_count, norm_count,
dot_count > norm_count * textord_noise_normratio
&& dot_count > 2 ? "REJECTED" : "ACCEPTED");
}
return super_norm_count < textord_noise_sncount
&& dot_count > norm_count * textord_noise_rowratio && dot_count > 2;
}
void Textord::clean_noise_from_words( //remove empties
ROW *row //row to clean
) {
TBOX blob_box; //bounding box
C_BLOB *blob; //current blob
C_OUTLINE *outline; //current outline
WERD *word; //current word
int32_t blob_size; //biggest size
int32_t trans_count; //no of transitions
int32_t trans_threshold; //noise tolerance
int32_t dot_count; //small objects
int32_t norm_count; //normal objects
int32_t dud_words; //number discarded
int32_t ok_words; //number remaining
int32_t word_index; //current word
//words of row
WERD_IT word_it = row->word_list ();
C_BLOB_IT blob_it; //blob iterator
C_OUTLINE_IT out_it; //outline iterator
ok_words = word_it.length ();
if (ok_words == 0 || textord_no_rejects)
return;
// was it chucked
std::vector<int8_t> word_dud(ok_words);
dud_words = 0;
ok_words = 0;
word_index = 0;
for (word_it.mark_cycle_pt (); !word_it.cycled_list (); word_it.forward ()) {
word = word_it.data (); //current word
dot_count = 0;
norm_count = 0;
//blobs in word
blob_it.set_to_list (word->cblob_list ());
for (blob_it.mark_cycle_pt (); !blob_it.cycled_list ();
blob_it.forward ()) {
blob = blob_it.data ();
if (!word->flag (W_DONT_CHOP)) {
//get outlines
out_it.set_to_list (blob->out_list ());
for (out_it.mark_cycle_pt (); !out_it.cycled_list ();
out_it.forward ()) {
outline = out_it.data ();
blob_box = outline->bounding_box ();
blob_size =
blob_box.width () >
blob_box.height ()? blob_box.width () : blob_box.
height();
if (blob_size < textord_noise_sizelimit * row->x_height ())
dot_count++; //count smal outlines
if (!outline->child ()->empty ()
&& blob_box.height () <
(1 + textord_noise_syfract) * row->x_height ()
&& blob_box.height () >
(1 - textord_noise_syfract) * row->x_height ()
&& blob_box.width () <
(1 + textord_noise_sxfract) * row->x_height ()
&& blob_box.width () >
(1 - textord_noise_sxfract) * row->x_height ())
norm_count++; //count smal outlines
}
}
else
norm_count++;
blob_box = blob->bounding_box ();
blob_size =
blob_box.width () >
blob_box.height ()? blob_box.width () : blob_box.height ();
if (blob_size >= textord_noise_sizelimit * row->x_height ()
&& blob_size < row->x_height () * 2) {
trans_threshold = blob_size / textord_noise_sizefraction;
trans_count = blob->count_transitions (trans_threshold);
if (trans_count < textord_noise_translimit)
norm_count++;
}
else if (blob_box.height () > row->x_height () * 2
&& (!word_it.at_first () || !blob_it.at_first ()))
dot_count += 2;
}
if (dot_count > 2 && !word->flag(W_REP_CHAR)) {
if (dot_count > norm_count * textord_noise_normratio * 2)
word_dud[word_index] = 2;
else if (dot_count > norm_count * textord_noise_normratio)
word_dud[word_index] = 1;
else
word_dud[word_index] = 0;
} else {
word_dud[word_index] = 0;
}
if (word_dud[word_index] == 2)
dud_words++;
else
ok_words++;
word_index++;
}
word_index = 0;
for (word_it.mark_cycle_pt (); !word_it.cycled_list (); word_it.forward ()) {
if (word_dud[word_index] == 2
|| (word_dud[word_index] == 1 && dud_words > ok_words)) {
word = word_it.data(); // Current word.
// Previously we threw away the entire word.
// Now just aggressively throw all small blobs into the reject list, where
// the classifier can decide whether they are actually needed.
word->CleanNoise(textord_noise_sizelimit * row->x_height());
}
word_index++;
}
}