// Insert the given blocks at the front of the completed_blocks_ list so
// they can be kept in the correct reading order.
void WorkingPartSet::InsertCompletedBlocks(BLOCK_LIST* blocks,
TO_BLOCK_LIST* to_blocks) {
BLOCK_IT block_it(&completed_blocks_);
block_it.add_list_before(blocks);
TO_BLOCK_IT to_block_it(&to_blocks_);
to_block_it.add_list_before(to_blocks);
}
// Make a block using lines parallel to the given vector that fit between
// the min and max coordinates specified by the ColPartitions.
// Construct a block from the given list of partitions.
void WorkingPartSet::MakeBlocks(const ICOORD& bleft, const ICOORD& tright,
int resolution, ColPartition_LIST* used_parts) {
part_it_.move_to_first();
while (!part_it_.empty()) {
// Gather a list of ColPartitions in block_parts that will be split
// by linespacing into smaller blocks.
ColPartition_LIST block_parts;
ColPartition_IT block_it(&block_parts);
ColPartition* next_part = NULL;
bool text_block = false;
do {
ColPartition* part = part_it_.extract();
if (part->blob_type() == BRT_UNKNOWN || part->blob_type() == BRT_TEXT)
text_block = true;
part->set_working_set(NULL);
part_it_.forward();
block_it.add_after_then_move(part);
next_part = part->SingletonPartner(false);
if (part_it_.empty() || next_part != part_it_.data()) {
// Sequences of partitions can get split by titles.
next_part = NULL;
}
// Merge adjacent blocks that are of the same type and let the
// linespacing determine the real boundaries.
if (next_part == NULL && !part_it_.empty()) {
ColPartition* next_block_part = part_it_.data();
const TBOX& part_box = part->bounding_box();
const TBOX& next_box = next_block_part->bounding_box();
// In addition to the same type, the next box must not be above the
// current box, nor (if image) too far below.
PolyBlockType type = part->type(), next_type = next_block_part->type();
if (ColPartition::TypesSimilar(type, next_type) &&
next_box.bottom() <= part_box.top() &&
(text_block ||
part_box.bottom() - next_box.top() < part_box.height()))
next_part = next_block_part;
}
} while (!part_it_.empty() && next_part != NULL);
if (!text_block) {
TO_BLOCK* to_block = ColPartition::MakeBlock(bleft, tright,
&block_parts, used_parts);
if (to_block != NULL) {
TO_BLOCK_IT to_block_it(&to_blocks_);
to_block_it.add_to_end(to_block);
BLOCK_IT block_it(&completed_blocks_);
block_it.add_to_end(to_block->block);
}
} else {
// Further sub-divide text blocks where linespacing changes.
ColPartition::LineSpacingBlocks(bleft, tright, resolution, &block_parts,
used_parts,
&completed_blocks_, &to_blocks_);
}
}
part_it_.set_to_list(&part_set_);
latest_part_ = NULL;
ASSERT_HOST(completed_blocks_.length() == to_blocks_.length());
}
/**
* Auto page segmentation. Divide the page image into blocks of uniform
* text linespacing and images.
*
* Resolution (in ppi) is derived from the input image.
*
* The output goes in the blocks list with corresponding TO_BLOCKs in the
* to_blocks list.
*
* If single_column is true, then no attempt is made to divide the image
* into columns, but multiple blocks are still made if the text is of
* non-uniform linespacing.
*
* If osd (orientation and script detection) is true then that is performed
* as well. If only_osd is true, then only orientation and script detection is
* performed. If osd is desired, (osd or only_osd) then osr_tess must be
* another Tesseract that was initialized especially for osd, and the results
* will be output into osr (orientation and script result).
*/
int Tesseract::AutoPageSeg(PageSegMode pageseg_mode,
BLOCK_LIST* blocks, TO_BLOCK_LIST* to_blocks,
Tesseract* osd_tess, OSResults* osr) {
if (textord_debug_images) {
WriteDebugBackgroundImage(textord_debug_printable, pix_binary_);
}
Pix* photomask_pix = NULL;
Pix* musicmask_pix = NULL;
// The blocks made by the ColumnFinder. Moved to blocks before return.
BLOCK_LIST found_blocks;
TO_BLOCK_LIST temp_blocks;
bool single_column = !PSM_COL_FIND_ENABLED(pageseg_mode);
bool osd_enabled = PSM_OSD_ENABLED(pageseg_mode);
bool osd_only = pageseg_mode == PSM_OSD_ONLY;
ColumnFinder* finder = SetupPageSegAndDetectOrientation(
single_column, osd_enabled, osd_only, blocks, osd_tess, osr,
&temp_blocks, &photomask_pix, &musicmask_pix);
int result = 0;
if (finder != NULL) {
TO_BLOCK_IT to_block_it(&temp_blocks);
TO_BLOCK* to_block = to_block_it.data();
if (musicmask_pix != NULL) {
// TODO(rays) pass the musicmask_pix into FindBlocks and mark music
// blocks separately. For now combine with photomask_pix.
pixOr(photomask_pix, photomask_pix, musicmask_pix);
}
if (equ_detect_) {
finder->SetEquationDetect(equ_detect_);
}
result = finder->FindBlocks(pageseg_mode, scaled_color_, scaled_factor_,
to_block, photomask_pix,
pix_thresholds_, pix_grey_,
&found_blocks, to_blocks);
if (result >= 0)
finder->GetDeskewVectors(&deskew_, &reskew_);
delete finder;
}
pixDestroy(&photomask_pix);
pixDestroy(&musicmask_pix);
if (result < 0) return result;
blocks->clear();
BLOCK_IT block_it(blocks);
// Move the found blocks to the input/output blocks.
block_it.add_list_after(&found_blocks);
if (textord_debug_images) {
// The debug image is no longer needed so delete it.
unlink(AlignedBlob::textord_debug_pix().string());
}
return result;
}
// Make blocks out of any partitions in this WorkingPartSet, and append
// them to the end of the blocks list. bleft, tright and resolution give
// the bounds and resolution of the source image, so that blocks can be
// made to fit in the bounds.
// All ColPartitions go in the used_parts list, as they need to be kept
// around, but are no longer needed.
void WorkingPartSet::ExtractCompletedBlocks(const ICOORD& bleft,
const ICOORD& tright,
int resolution,
ColPartition_LIST* used_parts,
BLOCK_LIST* blocks,
TO_BLOCK_LIST* to_blocks) {
MakeBlocks(bleft, tright, resolution, used_parts);
BLOCK_IT block_it(blocks);
block_it.move_to_last();
block_it.add_list_after(&completed_blocks_);
TO_BLOCK_IT to_block_it(to_blocks);
to_block_it.move_to_last();
to_block_it.add_list_after(&to_blocks_);
}
/**
* Auto page segmentation. Divide the page image into blocks of uniform
* text linespacing and images.
*
* Resolution (in ppi) is derived from the input image.
*
* The output goes in the blocks list with corresponding TO_BLOCKs in the
* to_blocks list.
*
* If !PSM_COL_FIND_ENABLED(pageseg_mode), then no attempt is made to divide
* the image into columns, but multiple blocks are still made if the text is
* of non-uniform linespacing.
*
* If diacritic_blobs is non-null, then diacritics/noise blobs, that would
* confuse layout anaylsis by causing textline overlap, are placed there,
* with the expectation that they will be reassigned to words later and
* noise/diacriticness determined via classification.
*
* If osd (orientation and script detection) is true then that is performed
* as well. If only_osd is true, then only orientation and script detection is
* performed. If osd is desired, (osd or only_osd) then osr_tess must be
* another Tesseract that was initialized especially for osd, and the results
* will be output into osr (orientation and script result).
*/
int Tesseract::AutoPageSeg(PageSegMode pageseg_mode, BLOCK_LIST* blocks,
TO_BLOCK_LIST* to_blocks,
BLOBNBOX_LIST* diacritic_blobs, Tesseract* osd_tess,
OSResults* osr) {
Pix* photomask_pix = NULL;
Pix* musicmask_pix = NULL;
// The blocks made by the ColumnFinder. Moved to blocks before return.
BLOCK_LIST found_blocks;
TO_BLOCK_LIST temp_blocks;
ColumnFinder* finder = SetupPageSegAndDetectOrientation(
pageseg_mode, blocks, osd_tess, osr, &temp_blocks, &photomask_pix,
&musicmask_pix);
int result = 0;
if (finder != NULL) {
TO_BLOCK_IT to_block_it(&temp_blocks);
TO_BLOCK* to_block = to_block_it.data();
if (musicmask_pix != NULL) {
// TODO(rays) pass the musicmask_pix into FindBlocks and mark music
// blocks separately. For now combine with photomask_pix.
pixOr(photomask_pix, photomask_pix, musicmask_pix);
}
if (equ_detect_) {
finder->SetEquationDetect(equ_detect_);
}
result = finder->FindBlocks(pageseg_mode, scaled_color_, scaled_factor_,
to_block, photomask_pix, pix_thresholds_,
pix_grey_, &pixa_debug_, &found_blocks,
diacritic_blobs, to_blocks);
if (result >= 0)
finder->GetDeskewVectors(&deskew_, &reskew_);
delete finder;
}
pixDestroy(&photomask_pix);
pixDestroy(&musicmask_pix);
if (result < 0) return result;
blocks->clear();
BLOCK_IT block_it(blocks);
// Move the found blocks to the input/output blocks.
block_it.add_list_after(&found_blocks);
return result;
}
/**
* Sets up auto page segmentation, determines the orientation, and corrects it.
* Somewhat arbitrary chunk of functionality, factored out of AutoPageSeg to
* facilitate testing.
* photo_mask_pix is a pointer to a NULL pointer that will be filled on return
* with the leptonica photo mask, which must be pixDestroyed by the caller.
* to_blocks is an empty list that will be filled with (usually a single)
* block that is used during layout analysis. This ugly API is required
* because of the possibility of a unlv zone file.
* TODO(rays) clean this up.
* See AutoPageSeg for other arguments.
* The returned ColumnFinder must be deleted after use.
*/
ColumnFinder* Tesseract::SetupPageSegAndDetectOrientation(
PageSegMode pageseg_mode, BLOCK_LIST* blocks, Tesseract* osd_tess,
OSResults* osr, TO_BLOCK_LIST* to_blocks, Pix** photo_mask_pix,
Pix** music_mask_pix) {
int vertical_x = 0;
int vertical_y = 1;
TabVector_LIST v_lines;
TabVector_LIST h_lines;
ICOORD bleft(0, 0);
ASSERT_HOST(pix_binary_ != NULL);
if (tessedit_dump_pageseg_images) {
pixa_debug_.AddPix(pix_binary_, "PageSegInput");
}
// Leptonica is used to find the rule/separator lines in the input.
LineFinder::FindAndRemoveLines(source_resolution_,
textord_tabfind_show_vlines, pix_binary_,
&vertical_x, &vertical_y, music_mask_pix,
&v_lines, &h_lines);
if (tessedit_dump_pageseg_images) {
pixa_debug_.AddPix(pix_binary_, "NoLines");
}
// Leptonica is used to find a mask of the photo regions in the input.
*photo_mask_pix = ImageFind::FindImages(pix_binary_, &pixa_debug_);
if (tessedit_dump_pageseg_images) {
pixa_debug_.AddPix(pix_binary_, "NoImages");
}
if (!PSM_COL_FIND_ENABLED(pageseg_mode)) v_lines.clear();
// The rest of the algorithm uses the usual connected components.
textord_.find_components(pix_binary_, blocks, to_blocks);
TO_BLOCK_IT to_block_it(to_blocks);
// There must be exactly one input block.
// TODO(rays) handle new textline finding with a UNLV zone file.
ASSERT_HOST(to_blocks->singleton());
TO_BLOCK* to_block = to_block_it.data();
TBOX blkbox = to_block->block->bounding_box();
ColumnFinder* finder = NULL;
int estimated_resolution = source_resolution_;
if (source_resolution_ == kMinCredibleResolution) {
// Try to estimate resolution from typical body text size.
int res = IntCastRounded(to_block->line_size * kResolutionEstimationFactor);
if (res > estimated_resolution && res < kMaxCredibleResolution) {
estimated_resolution = res;
tprintf("Estimating resolution as %d\n", estimated_resolution);
}
}
if (to_block->line_size >= 2) {
finder = new ColumnFinder(static_cast<int>(to_block->line_size),
blkbox.botleft(), blkbox.topright(),
estimated_resolution, textord_use_cjk_fp_model,
textord_tabfind_aligned_gap_fraction, &v_lines,
&h_lines, vertical_x, vertical_y);
finder->SetupAndFilterNoise(pageseg_mode, *photo_mask_pix, to_block);
if (equ_detect_) {
equ_detect_->LabelSpecialText(to_block);
}
BLOBNBOX_CLIST osd_blobs;
// osd_orientation is the number of 90 degree rotations to make the
// characters upright. (See osdetect.h for precise definition.)
// We want the text lines horizontal, (vertical text indicates vertical
// textlines) which may conflict (eg vertically written CJK).
int osd_orientation = 0;
bool vertical_text = textord_tabfind_force_vertical_text ||
pageseg_mode == PSM_SINGLE_BLOCK_VERT_TEXT;
if (!vertical_text && textord_tabfind_vertical_text &&
PSM_ORIENTATION_ENABLED(pageseg_mode)) {
vertical_text =
finder->IsVerticallyAlignedText(textord_tabfind_vertical_text_ratio,
to_block, &osd_blobs);
}
if (PSM_OSD_ENABLED(pageseg_mode) && osd_tess != NULL && osr != NULL) {
GenericVector<int> osd_scripts;
if (osd_tess != this) {
// We are running osd as part of layout analysis, so constrain the
// scripts to those allowed by *this.
AddAllScriptsConverted(unicharset, osd_tess->unicharset, &osd_scripts);
for (int s = 0; s < sub_langs_.size(); ++s) {
AddAllScriptsConverted(sub_langs_[s]->unicharset,
osd_tess->unicharset, &osd_scripts);
}
}
os_detect_blobs(&osd_scripts, &osd_blobs, osr, osd_tess);
if (pageseg_mode == PSM_OSD_ONLY) {
delete finder;
return NULL;
}
osd_orientation = osr->best_result.orientation_id;
double osd_score = osr->orientations[osd_orientation];
double osd_margin = min_orientation_margin * 2;
for (int i = 0; i < 4; ++i) {
if (i != osd_orientation &&
osd_score - osr->orientations[i] < osd_margin) {
osd_margin = osd_score - osr->orientations[i];
}
}
int best_script_id = osr->best_result.script_id;
const char* best_script_str =
osd_tess->unicharset.get_script_from_script_id(best_script_id);
bool cjk = best_script_id == osd_tess->unicharset.han_sid() ||
best_script_id == osd_tess->unicharset.hiragana_sid() ||
best_script_id == osd_tess->unicharset.katakana_sid() ||
strcmp("Japanese", best_script_str) == 0 ||
strcmp("Korean", best_script_str) == 0 ||
strcmp("Hangul", best_script_str) == 0;
if (cjk) {
finder->set_cjk_script(true);
}
if (osd_margin < min_orientation_margin) {
// The margin is weak.
if (!cjk && !vertical_text && osd_orientation == 2) {
// upside down latin text is improbable with such a weak margin.
tprintf("OSD: Weak margin (%.2f), horiz textlines, not CJK: "
"Don't rotate.\n", osd_margin);
osd_orientation = 0;
} else {
tprintf(
"OSD: Weak margin (%.2f) for %d blob text block, "
"but using orientation anyway: %d\n",
osd_margin, osd_blobs.length(), osd_orientation);
}
}
}
osd_blobs.shallow_clear();
finder->CorrectOrientation(to_block, vertical_text, osd_orientation);
}
return finder;
}
// Make the textlines and words inside each block.
void Textord::TextordPage(PageSegMode pageseg_mode, const FCOORD &reskew,
int width, int height, Pix *binary_pix,
Pix *thresholds_pix, Pix *grey_pix,
bool use_box_bottoms,
BLOCK_LIST *blocks, TO_BLOCK_LIST *to_blocks) {
page_tr_.set_x(width);
page_tr_.set_y(height);
if (to_blocks->empty()) {
// AutoPageSeg was not used, so we need to find_components first.
find_components(binary_pix, blocks, to_blocks);
TO_BLOCK_IT it(to_blocks);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
TO_BLOCK *to_block = it.data();
// Compute the edge offsets whether or not there is a grey_pix.
// We have by-passed auto page seg, so we have to run it here.
// By page segmentation mode there is no non-text to avoid running on.
to_block->ComputeEdgeOffsets(thresholds_pix, grey_pix);
}
} else if (!PSM_SPARSE(pageseg_mode)) {
// AutoPageSeg does not need to find_components as it did that already.
// Filter_blobs sets up the TO_BLOCKs the same as find_components does.
filter_blobs(page_tr_, to_blocks, true);
}
ASSERT_HOST(!to_blocks->empty());
if (pageseg_mode == PSM_SINGLE_BLOCK_VERT_TEXT) {
const FCOORD anticlockwise90(0.0f, 1.0f);
const FCOORD clockwise90(0.0f, -1.0f);
TO_BLOCK_IT it(to_blocks);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
TO_BLOCK *to_block = it.data();
BLOCK *block = to_block->block;
// Create a fake poly_block in block from its bounding box.
block->set_poly_block(new POLY_BLOCK(block->bounding_box(),
PT_VERTICAL_TEXT));
// Rotate the to_block along with its contained block and blobnbox lists.
to_block->rotate(anticlockwise90);
// Set the block's rotation values to obey the convention followed in
// layout analysis for vertical text.
block->set_re_rotation(clockwise90);
block->set_classify_rotation(clockwise90);
}
}
TO_BLOCK_IT to_block_it(to_blocks);
TO_BLOCK *to_block = to_block_it.data();
// Make the rows in the block.
float gradient = 0;
// Do it the old fashioned way.
if (PSM_LINE_FIND_ENABLED(pageseg_mode)) {
gradient = make_rows(page_tr_, to_blocks);
} else if (!PSM_SPARSE(pageseg_mode)) {
// RAW_LINE, SINGLE_LINE, SINGLE_WORD and SINGLE_CHAR all need a single row.
gradient = make_single_row(page_tr_, pageseg_mode != PSM_RAW_LINE,
to_block, to_blocks);
}
BaselineDetect baseline_detector(textord_baseline_debug,
reskew, to_blocks);
baseline_detector.ComputeStraightBaselines(use_box_bottoms);
baseline_detector.ComputeBaselineSplinesAndXheights(page_tr_, true,
textord_heavy_nr,
textord_show_final_rows,
this);
// Now make the words in the lines.
if (PSM_WORD_FIND_ENABLED(pageseg_mode)) {
// SINGLE_LINE uses the old word maker on the single line.
make_words(this, page_tr_, gradient, blocks, to_blocks);
} else {
// SINGLE_WORD and SINGLE_CHAR cram all the blobs into a
// single word, and in SINGLE_CHAR mode, all the outlines
// go in a single blob.
TO_BLOCK *to_block = to_block_it.data();
make_single_word(pageseg_mode == PSM_SINGLE_CHAR,
to_block->get_rows(), to_block->block->row_list());
}
cleanup_blocks(PSM_WORD_FIND_ENABLED(pageseg_mode), blocks);
// Remove empties.
// Compute the margins for each row in the block, to be used later for
// paragraph detection.
BLOCK_IT b_it(blocks);
for (b_it.mark_cycle_pt(); !b_it.cycled_list(); b_it.forward()) {
b_it.data()->compute_row_margins();
}
#ifndef GRAPHICS_DISABLED
close_to_win();
#endif
}
/**
* Sets up auto page segmentation, determines the orientation, and corrects it.
* Somewhat arbitrary chunk of functionality, factored out of AutoPageSeg to
* facilitate testing.
* photo_mask_pix is a pointer to a NULL pointer that will be filled on return
* with the leptonica photo mask, which must be pixDestroyed by the caller.
* to_blocks is an empty list that will be filled with (usually a single)
* block that is used during layout analysis. This ugly API is required
* because of the possibility of a unlv zone file.
* TODO(rays) clean this up.
* See AutoPageSeg for other arguments.
* The returned ColumnFinder must be deleted after use.
*/
ColumnFinder* Tesseract::SetupPageSegAndDetectOrientation(
bool single_column, bool osd, bool only_osd,
BLOCK_LIST* blocks, Tesseract* osd_tess, OSResults* osr,
TO_BLOCK_LIST* to_blocks, Pix** photo_mask_pix, Pix** music_mask_pix) {
int vertical_x = 0;
int vertical_y = 1;
TabVector_LIST v_lines;
TabVector_LIST h_lines;
ICOORD bleft(0, 0);
ASSERT_HOST(pix_binary_ != NULL);
if (tessedit_dump_pageseg_images) {
pixWrite("tessinput.png", pix_binary_, IFF_PNG);
}
// Leptonica is used to find the rule/separator lines in the input.
LineFinder::FindAndRemoveLines(source_resolution_,
textord_tabfind_show_vlines, pix_binary_,
&vertical_x, &vertical_y, music_mask_pix,
&v_lines, &h_lines);
if (tessedit_dump_pageseg_images)
pixWrite("tessnolines.png", pix_binary_, IFF_PNG);
// Leptonica is used to find a mask of the photo regions in the input.
*photo_mask_pix = ImageFind::FindImages(pix_binary_);
if (tessedit_dump_pageseg_images)
pixWrite("tessnoimages.png", pix_binary_, IFF_PNG);
if (single_column)
v_lines.clear();
// The rest of the algorithm uses the usual connected components.
textord_.find_components(pix_binary_, blocks, to_blocks);
TO_BLOCK_IT to_block_it(to_blocks);
// There must be exactly one input block.
// TODO(rays) handle new textline finding with a UNLV zone file.
ASSERT_HOST(to_blocks->singleton());
TO_BLOCK* to_block = to_block_it.data();
TBOX blkbox = to_block->block->bounding_box();
ColumnFinder* finder = NULL;
if (to_block->line_size >= 2) {
finder = new ColumnFinder(static_cast<int>(to_block->line_size),
blkbox.botleft(), blkbox.topright(),
source_resolution_,
&v_lines, &h_lines, vertical_x, vertical_y);
finder->SetupAndFilterNoise(*photo_mask_pix, to_block);
if (equ_detect_) {
equ_detect_->LabelSpecialText(to_block);
}
BLOBNBOX_CLIST osd_blobs;
// osd_orientation is the number of 90 degree rotations to make the
// characters upright. (See osdetect.h for precise definition.)
// We want the text lines horizontal, (vertical text indicates vertical
// textlines) which may conflict (eg vertically written CJK).
int osd_orientation = 0;
bool vertical_text = finder->IsVerticallyAlignedText(to_block, &osd_blobs);
if (osd && osd_tess != NULL && osr != NULL) {
os_detect_blobs(&osd_blobs, osr, osd_tess);
if (only_osd) {
delete finder;
return NULL;
}
osd_orientation = osr->best_result.orientation_id;
double osd_score = osr->orientations[osd_orientation];
double osd_margin = min_orientation_margin * 2;
for (int i = 0; i < 4; ++i) {
if (i != osd_orientation &&
osd_score - osr->orientations[i] < osd_margin) {
osd_margin = osd_score - osr->orientations[i];
}
}
if (osd_margin < min_orientation_margin) {
// The margin is weak.
int best_script_id = osr->best_result.script_id;
bool cjk = (best_script_id == osd_tess->unicharset.han_sid()) ||
(best_script_id == osd_tess->unicharset.hiragana_sid()) ||
(best_script_id == osd_tess->unicharset.katakana_sid());
if (!cjk && !vertical_text && osd_orientation == 2) {
// upside down latin text is improbable with such a weak margin.
tprintf("OSD: Weak margin (%.2f), horiz textlines, not CJK: "
"Don't rotate.\n", osd_margin);
osd_orientation = 0;
} else {
tprintf("OSD: Weak margin (%.2f) for %d blob text block, "
"but using orientation anyway: %d\n",
osd_blobs.length(), osd_margin, osd_orientation);
}
}
}
osd_blobs.shallow_clear();
finder->CorrectOrientation(to_block, vertical_text, osd_orientation);
}
return finder;
}
// Segment the page according to the current value of tessedit_pageseg_mode.
// If the pix_binary_ member is not NULL, it is used as the source image,
// and copied to image, otherwise it just uses image as the input.
// On return the blocks list owns all the constructed page layout.
int Tesseract::SegmentPage(const STRING* input_file,
IMAGE* image, BLOCK_LIST* blocks) {
int width = image->get_xsize();
int height = image->get_ysize();
int resolution = image->get_res();
#ifdef HAVE_LIBLEPT
if (pix_binary_ != NULL) {
width = pixGetWidth(pix_binary_);
height = pixGetHeight(pix_binary_);
resolution = pixGetXRes(pix_binary_);
}
#endif
// Zero resolution messes up the algorithms, so make sure it is credible.
if (resolution < kMinCredibleResolution)
resolution = kDefaultResolution;
// Get page segmentation mode.
PageSegMode pageseg_mode = static_cast<PageSegMode>(
static_cast<int>(tessedit_pageseg_mode));
// If a UNLV zone file can be found, use that instead of segmentation.
if (pageseg_mode != tesseract::PSM_AUTO &&
input_file != NULL && input_file->length() > 0) {
STRING name = *input_file;
const char* lastdot = strrchr(name.string(), '.');
if (lastdot != NULL)
name[lastdot - name.string()] = '\0';
read_unlv_file(name, width, height, blocks);
}
bool single_column = pageseg_mode > PSM_AUTO;
if (blocks->empty()) {
// No UNLV file present. Work according to the PageSegMode.
// First make a single block covering the whole image.
BLOCK_IT block_it(blocks);
BLOCK* block = new BLOCK("", TRUE, 0, 0, 0, 0, width, height);
block_it.add_to_end(block);
} else {
// UNLV file present. Use PSM_SINGLE_COLUMN.
pageseg_mode = PSM_SINGLE_COLUMN;
}
TO_BLOCK_LIST land_blocks, port_blocks;
TBOX page_box;
if (pageseg_mode <= PSM_SINGLE_COLUMN) {
if (AutoPageSeg(width, height, resolution, single_column,
image, blocks, &port_blocks) < 0) {
return -1;
}
// To create blobs from the image region bounds uncomment this line:
// port_blocks.clear(); // Uncomment to go back to the old mode.
} else {
#if HAVE_LIBLEPT
image->FromPix(pix_binary_);
#endif
deskew_ = FCOORD(1.0f, 0.0f);
reskew_ = FCOORD(1.0f, 0.0f);
}
if (blocks->empty()) {
tprintf("Empty page\n");
return 0; // AutoPageSeg found an empty page.
}
if (port_blocks.empty()) {
// AutoPageSeg was not used, so we need to find_components first.
find_components(blocks, &land_blocks, &port_blocks, &page_box);
} else {
// AutoPageSeg does not need to find_components as it did that already.
page_box.set_left(0);
page_box.set_bottom(0);
page_box.set_right(width);
page_box.set_top(height);
// Filter_blobs sets up the TO_BLOCKs the same as find_components does.
filter_blobs(page_box.topright(), &port_blocks, true);
}
TO_BLOCK_IT to_block_it(&port_blocks);
ASSERT_HOST(!port_blocks.empty());
TO_BLOCK* to_block = to_block_it.data();
if (pageseg_mode <= PSM_SINGLE_BLOCK ||
to_block->line_size < 2) {
// For now, AUTO, SINGLE_COLUMN and SINGLE_BLOCK all map to the old
// textord. The difference is the number of blocks and how the are made.
textord_page(page_box.topright(), blocks, &land_blocks, &port_blocks,
this);
} else {
// SINGLE_LINE, SINGLE_WORD and SINGLE_CHAR all need a single row.
float gradient = make_single_row(page_box.topright(),
to_block, &port_blocks, this);
if (pageseg_mode == PSM_SINGLE_LINE) {
// SINGLE_LINE uses the old word maker on the single line.
make_words(page_box.topright(), gradient, blocks,
&land_blocks, &port_blocks, this);
} else {
// SINGLE_WORD and SINGLE_CHAR cram all the blobs into a
// single word, and in SINGLE_CHAR mode, all the outlines
// go in a single blob.
make_single_word(pageseg_mode == PSM_SINGLE_CHAR,
to_block->get_rows(), to_block->block->row_list());
}
}
return 0;
}
// Auto page segmentation. Divide the page image into blocks of uniform
// text linespacing and images.
// Width, height and resolution are derived from the input image.
// If the pix is non-NULL, then it is assumed to be the input, and it is
// copied to the image, otherwise the image is used directly.
// The output goes in the blocks list with corresponding TO_BLOCKs in the
// to_blocks list.
// If single_column is true, then no attempt is made to divide the image
// into columns, but multiple blocks are still made if the text is of
// non-uniform linespacing.
int Tesseract::AutoPageSeg(int width, int height, int resolution,
bool single_column, IMAGE* image,
BLOCK_LIST* blocks, TO_BLOCK_LIST* to_blocks) {
int vertical_x = 0;
int vertical_y = 1;
TabVector_LIST v_lines;
TabVector_LIST h_lines;
ICOORD bleft(0, 0);
Boxa* boxa = NULL;
Pixa* pixa = NULL;
// The blocks made by the ColumnFinder. Moved to blocks before return.
BLOCK_LIST found_blocks;
#ifdef HAVE_LIBLEPT
if (pix_binary_ != NULL) {
if (textord_debug_images) {
Pix* grey_pix = pixCreate(width, height, 8);
// Printable images are light grey on white, but for screen display
// they are black on dark grey so the other colors show up well.
if (textord_debug_printable) {
pixSetAll(grey_pix);
pixSetMasked(grey_pix, pix_binary_, 192);
} else {
pixSetAllArbitrary(grey_pix, 64);
pixSetMasked(grey_pix, pix_binary_, 0);
}
AlignedBlob::IncrementDebugPix();
pixWrite(AlignedBlob::textord_debug_pix().string(), grey_pix, IFF_PNG);
pixDestroy(&grey_pix);
}
if (tessedit_dump_pageseg_images)
pixWrite("tessinput.png", pix_binary_, IFF_PNG);
// Leptonica is used to find the lines and image regions in the input.
LineFinder::FindVerticalLines(resolution, pix_binary_,
&vertical_x, &vertical_y, &v_lines);
LineFinder::FindHorizontalLines(resolution, pix_binary_, &h_lines);
if (tessedit_dump_pageseg_images)
pixWrite("tessnolines.png", pix_binary_, IFF_PNG);
ImageFinder::FindImages(pix_binary_, &boxa, &pixa);
if (tessedit_dump_pageseg_images)
pixWrite("tessnoimages.png", pix_binary_, IFF_PNG);
// Copy the Pix to the IMAGE.
image->FromPix(pix_binary_);
if (single_column)
v_lines.clear();
}
#endif
TO_BLOCK_LIST land_blocks, port_blocks;
TBOX page_box;
// The rest of the algorithm uses the usual connected components.
find_components(blocks, &land_blocks, &port_blocks, &page_box);
TO_BLOCK_IT to_block_it(&port_blocks);
ASSERT_HOST(!to_block_it.empty());
for (to_block_it.mark_cycle_pt(); !to_block_it.cycled_list();
to_block_it.forward()) {
TO_BLOCK* to_block = to_block_it.data();
TBOX blkbox = to_block->block->bounding_box();
if (to_block->line_size >= 2) {
// Note: if there are multiple blocks, then v_lines, boxa, and pixa
// are empty on the next iteration, but in this case, we assume
// that there aren't any interesting line separators or images, since
// it means that we have a pre-defined unlv zone file.
ColumnFinder finder(static_cast<int>(to_block->line_size),
blkbox.botleft(), blkbox.topright(),
&v_lines, &h_lines, vertical_x, vertical_y);
if (finder.FindBlocks(height, resolution, single_column,
to_block, boxa, pixa, &found_blocks, to_blocks) < 0)
return -1;
finder.ComputeDeskewVectors(&deskew_, &reskew_);
boxa = NULL;
pixa = NULL;
}
}
#ifdef HAVE_LIBLEPT
boxaDestroy(&boxa);
pixaDestroy(&pixa);
#endif
blocks->clear();
BLOCK_IT block_it(blocks);
// Move the found blocks to the input/output blocks.
block_it.add_list_after(&found_blocks);
if (textord_debug_images) {
// The debug image is no longer needed so delete it.
unlink(AlignedBlob::textord_debug_pix().string());
}
return 0;
}
