void Tesseract::PrerecAllWordsPar(const GenericVector<WordData>& words) {
// Prepare all the blobs.
GenericVector<BlobData> blobs;
for (int w = 0; w < words.size(); ++w) {
if (words[w].word->ratings != NULL &&
words[w].word->ratings->get(0, 0) == NULL) {
for (int b = 0; b < words[w].word->chopped_word->NumBlobs(); ++b) {
blobs.push_back(BlobData(b, this, *words[w].word));
}
for (int s = 0; s < words[w].lang_words.size(); ++s) {
const WERD_RES& word = words[w].lang_words[s];
for (int b = 0; b < word.chopped_word->NumBlobs(); ++b) {
blobs.push_back(BlobData(b, sub_langs_[s], word));
}
}
}
}
// Pre-classify all the blobs.
if (tessedit_parallelize > 1) {
#pragma omp parallel for num_threads(10)
for (int b = 0; b < blobs.size(); ++b) {
*blobs[b].choices =
blobs[b].tesseract->classify_blob(blobs[b].blob, "par", White, NULL);
}
} else {
// TODO(AMD) parallelize this.
for (int b = 0; b < blobs.size(); ++b) {
*blobs[b].choices =
blobs[b].tesseract->classify_blob(blobs[b].blob, "par", White, NULL);
}
}
}
// Given an initial estimate of line spacing (m_in) and the positions of each
// baseline, computes the line spacing of the block more accurately in m_out,
// and the corresponding intercept in c_out, and the number of spacings seen
// in index_delta. Returns the error of fit to the line spacing model.
// Uses a simple linear regression, but optimized the offset using the median.
double BaselineBlock::FitLineSpacingModel(
const GenericVector<double>& positions, double m_in,
double* m_out, double* c_out, int* index_delta) {
if (m_in == 0.0f || positions.size() < 2) {
*m_out = m_in;
*c_out = 0.0;
if (index_delta != NULL) *index_delta = 0;
return 0.0;
}
GenericVector<double> offsets;
// Get the offset (remainder) linespacing for each line and choose the median.
for (int i = 0; i < positions.size(); ++i)
offsets.push_back(fmod(positions[i], m_in));
// Get the median offset.
double median_offset = MedianOfCircularValues(m_in, &offsets);
// Now fit a line to quantized line number and offset.
LLSQ llsq;
int min_index = MAX_INT32;
int max_index = -MAX_INT32;
for (int i = 0; i < positions.size(); ++i) {
double y_pos = positions[i];
int row_index = IntCastRounded((y_pos - median_offset) / m_in);
UpdateRange(row_index, &min_index, &max_index);
llsq.add(row_index, y_pos);
}
// Get the refined line spacing.
*m_out = llsq.m();
// Use the median offset rather than the mean.
offsets.truncate(0);
for (int i = 0; i < positions.size(); ++i)
offsets.push_back(fmod(positions[i], *m_out));
// Get the median offset.
if (debug_level_ > 2) {
for (int i = 0; i < offsets.size(); ++i)
tprintf("%d: %g\n", i, offsets[i]);
}
*c_out = MedianOfCircularValues(*m_out, &offsets);
if (debug_level_ > 1) {
tprintf("Median offset = %g, compared to mean of %g.\n",
*c_out, llsq.c(*m_out));
}
// Index_delta is the number of hypothesized line gaps present.
if (index_delta != NULL)
*index_delta = max_index - min_index;
// Use the regression model's intercept to compute the error, as it may be
// a full line-spacing in disagreement with the median.
double rms_error = llsq.rms(*m_out, llsq.c(*m_out));
if (debug_level_ > 1) {
tprintf("Linespacing of y=%g x + %g improved to %g x + %g, rms=%g\n",
m_in, median_offset, *m_out, *c_out, rms_error);
}
return rms_error;
}
// Parse a string of the form [~]<lang>[+[~]<lang>]*.
// Langs with no prefix get appended to to_load, provided they
// are not in there already.
// Langs with ~ prefix get appended to not_to_load, provided they are not in
// there already.
void Tesseract::ParseLanguageString(const char* lang_str,
GenericVector<STRING>* to_load,
GenericVector<STRING>* not_to_load) {
STRING remains(lang_str);
while (remains.length() > 0) {
// Find the start of the lang code and which vector to add to.
const char* start = remains.string();
while (*start == '+')
++start;
GenericVector<STRING>* target = to_load;
if (*start == '~') {
target = not_to_load;
++start;
}
// Find the index of the end of the lang code in string start.
int end = strlen(start);
const char* plus = strchr(start, '+');
if (plus != NULL && plus - start < end)
end = plus - start;
STRING lang_code(start);
lang_code.truncate_at(end);
STRING next(start + end);
remains = next;
// Check whether lang_code is already in the target vector and add.
if (!IsStrInList(lang_code, *target)) {
if (tessdata_manager_debug_level)
tprintf("Adding language '%s' to list\n", lang_code.string());
target->push_back(lang_code);
}
}
}
// Computes an estimate of the line spacing of the block from the median
// of the spacings between adjacent overlapping textlines.
void BaselineBlock::EstimateLineSpacing() {
GenericVector<float> spacings;
for (int r = 0; r < rows_.size(); ++r) {
BaselineRow* row = rows_[r];
// Exclude silly lines.
if (fabs(row->BaselineAngle()) > M_PI * 0.25) continue;
// Find the first row after row that overlaps it significantly.
const TBOX& row_box = row->bounding_box();
int r2;
for (r2 = r + 1; r2 < rows_.size() &&
!row_box.major_x_overlap(rows_[r2]->bounding_box());
++r2);
if (r2 < rows_.size()) {
BaselineRow* row2 = rows_[r2];
// Exclude silly lines.
if (fabs(row2->BaselineAngle()) > M_PI * 0.25) continue;
float spacing = row->SpaceBetween(*row2);
spacings.push_back(spacing);
}
}
// If we have at least one value, use it, otherwise leave the previous
// value unchanged.
if (!spacings.empty()) {
line_spacing_ = spacings[spacings.choose_nth_item(spacings.size() / 2)];
if (debug_level_ > 1)
tprintf("Estimate of linespacing = %g\n", line_spacing_);
}
}
// Fits straight line baselines and computes the skew angle from the
// median angle. Returns true if a good angle is found.
// If use_box_bottoms is false, baseline positions are formed by
// considering the outlines of the blobs.
bool BaselineBlock::FitBaselinesAndFindSkew(bool use_box_bottoms) {
if (non_text_block_) return false;
GenericVector<double> angles;
for (int r = 0; r < rows_.size(); ++r) {
BaselineRow* row = rows_[r];
if (row->FitBaseline(use_box_bottoms)) {
double angle = row->BaselineAngle();
angles.push_back(angle);
}
if (debug_level_ > 1)
row->Print();
}
if (!angles.empty()) {
skew_angle_ = MedianOfCircularValues(M_PI, &angles);
good_skew_angle_ = true;
} else {
skew_angle_ = 0.0f;
good_skew_angle_ = false;
}
if (debug_level_ > 0) {
tprintf("Initial block skew angle = %g, good = %d\n",
skew_angle_, good_skew_angle_);
}
return good_skew_angle_;
}
TESS_API int TESS_CALL TessBaseAPIInit4(TessBaseAPI* handle, const char* datapath, const char* language,
TessOcrEngineMode mode, char** configs, int configs_size,
char** vars_vec, char** vars_values, size_t vars_vec_size,
BOOL set_only_non_debug_params)
{
GenericVector<STRING> varNames;
GenericVector<STRING> varValues;
if (vars_vec != nullptr && vars_values != nullptr) {
for (size_t i = 0; i < vars_vec_size; i++) {
varNames.push_back(STRING(vars_vec[i]));
varValues.push_back(STRING(vars_values[i]));
}
}
return handle->Init(datapath, language, mode, configs, configs_size, &varNames, &varValues, set_only_non_debug_params);
}
// Applies the box file based on the image name fname, and resegments
// the words in the block_list (page), with:
// blob-mode: one blob per line in the box file, words as input.
// word/line-mode: one blob per space-delimited unit after the #, and one word
// per line in the box file. (See comment above for box file format.)
// If find_segmentation is true, (word/line mode) then the classifier is used
// to re-segment words/lines to match the space-delimited truth string for
// each box. In this case, the input box may be for a word or even a whole
// text line, and the output words will contain multiple blobs corresponding
// to the space-delimited input string.
// With find_segmentation false, no classifier is needed, but the chopper
// can still be used to correctly segment touching characters with the help
// of the input boxes.
// In the returned PAGE_RES, the WERD_RES are setup as they would be returned
// from normal classification, ie. with a word, chopped_word, rebuild_word,
// seam_array, denorm, box_word, and best_state, but NO best_choice or
// raw_choice, as they would require a UNICHARSET, which we aim to avoid.
// Instead, the correct_text member of WERD_RES is set, and this may be later
// converted to a best_choice using CorrectClassifyWords. CorrectClassifyWords
// is not required before calling ApplyBoxTraining.
PAGE_RES* Tesseract::ApplyBoxes(const STRING& fname,
bool find_segmentation,
BLOCK_LIST *block_list) {
GenericVector<TBOX> boxes;
GenericVector<STRING> texts, full_texts;
if (!ReadAllBoxes(applybox_page, true, fname, &boxes, &texts, &full_texts,
NULL)) {
return NULL; // Can't do it.
}
int box_count = boxes.size();
int box_failures = 0;
// Add an empty everything to the end.
boxes.push_back(TBOX());
texts.push_back(STRING());
full_texts.push_back(STRING());
// In word mode, we use the boxes to make a word for each box, but
// in blob mode we use the existing words and maximally chop them first.
PAGE_RES* page_res = find_segmentation ?
NULL : SetupApplyBoxes(boxes, block_list);
clear_any_old_text(block_list);
for (int i = 0; i < boxes.size() - 1; i++) {
bool foundit = false;
if (page_res != NULL) {
if (i == 0) {
foundit = ResegmentCharBox(page_res, NULL, boxes[i], boxes[i + 1],
full_texts[i].string());
} else {
foundit = ResegmentCharBox(page_res, &boxes[i-1], boxes[i],
boxes[i + 1], full_texts[i].string());
}
} else {
foundit = ResegmentWordBox(block_list, boxes[i], boxes[i + 1],
texts[i].string());
}
if (!foundit) {
box_failures++;
ReportFailedBox(i, boxes[i], texts[i].string(),
"FAILURE! Couldn't find a matching blob");
}
}
if (page_res == NULL) {
// In word/line mode, we now maximally chop all the words and resegment
// them with the classifier.
page_res = SetupApplyBoxes(boxes, block_list);
ReSegmentByClassification(page_res);
}
if (applybox_debug > 0) {
tprintf("APPLY_BOXES:\n");
tprintf(" Boxes read from boxfile: %6d\n", box_count);
if (box_failures > 0)
tprintf(" Boxes failed resegmentation: %6d\n", box_failures);
}
TidyUp(page_res);
return page_res;
}
GenericVector<char*> M_Utils::lineSplit(const char* txt) {
int txtlen = (int)strlen(txt);
// pass 1: find split points
GenericVector<int> splitpoints;
for(int i = 0; i < txtlen; i++) {
if(txt[i] == '\n' && (i < (txtlen-1)))
splitpoints.push_back(i);
}
// pass 2: iterate split points to do all the splitting
int prevsplit = 0;
GenericVector<char*> res;
if(splitpoints.empty()) {
// deep copy the string
char* newstr = strDeepCpy(txt);
res.push_back(newstr);
return res;
}
for(int i = 0; i < splitpoints.length(); i++) {
int split = splitpoints[i];
int newstrsize = split-prevsplit;
char* ln = new char[newstrsize+2]; // +1 for null terminator and +1 for newline
for(int i = 0; i < newstrsize; i++)
ln[i] = txt[prevsplit+i];
ln[newstrsize] = '\n';
ln[newstrsize+1] = '\0'; // null terminator
res.push_back(ln);
splitpoints.clear();
prevsplit = split;
}
// now just need to add the last line
int lastsplit = prevsplit;
int newstrsize = txtlen - prevsplit;
char* ln = new char[newstrsize+1];
for(int i = 0; i < newstrsize; i++)
ln[i] = txt[prevsplit+i];
ln[newstrsize] = '\0';
res.push_back(ln);
return res;
}
/**
* WERD_CHOICE::WERD_CHOICE
*
* Constructor to build a WERD_CHOICE from the given string.
* The function assumes that src_string is not NULL.
*/
WERD_CHOICE::WERD_CHOICE(const char *src_string,
const UNICHARSET &unicharset)
: unicharset_(&unicharset){
GenericVector<UNICHAR_ID> encoding;
GenericVector<char> lengths;
if (unicharset.encode_string(src_string, true, &encoding, &lengths, NULL)) {
lengths.push_back('\0');
STRING src_lengths = &lengths[0];
this->init(src_string, src_lengths.string(), 0.0, 0.0, NO_PERM);
} else { // There must have been an invalid unichar in the string.
this->init(8);
this->make_bad();
}
}
void ResultIterator::CalculateTextlineOrder(
bool paragraph_is_ltr,
const LTRResultIterator &resit,
GenericVector<StrongScriptDirection> *dirs_arg,
GenericVectorEqEq<int> *word_indices) const {
GenericVector<StrongScriptDirection> dirs;
GenericVector<StrongScriptDirection> *directions;
directions = (dirs_arg != NULL) ? dirs_arg : &dirs;
directions->truncate(0);
// A LTRResultIterator goes strictly left-to-right word order.
LTRResultIterator ltr_it(resit);
ltr_it.RestartRow();
if (ltr_it.Empty(RIL_WORD)) return;
do {
directions->push_back(ltr_it.WordDirection());
} while (ltr_it.Next(RIL_WORD) && !ltr_it.IsAtBeginningOf(RIL_TEXTLINE));
word_indices->truncate(0);
CalculateTextlineOrder(paragraph_is_ltr, *directions, word_indices);
}