/**
* 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;
}
/**
* 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;
}
/**
* 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;
}
