|
Tesseract
3.02
|
|
Tesseract
3.02
|
#include <colfind.h>
Public Member Functions | |
| ColumnFinder (int gridsize, const ICOORD &bleft, const ICOORD &tright, int resolution, TabVector_LIST *vlines, TabVector_LIST *hlines, int vertical_x, int vertical_y) | |
| virtual | ~ColumnFinder () |
| const DENORM * | denorm () const |
| const TextlineProjection * | projection () const |
| void | SetupAndFilterNoise (Pix *photo_mask_pix, TO_BLOCK *input_block) |
| bool | IsVerticallyAlignedText (TO_BLOCK *block, BLOBNBOX_CLIST *osd_blobs) |
| void | CorrectOrientation (TO_BLOCK *block, bool vertical_text_lines, int recognition_rotation) |
| int | FindBlocks (bool single_column, Pix *scaled_color, int scaled_factor, TO_BLOCK *block, Pix *photo_mask_pix, BLOCK_LIST *blocks, TO_BLOCK_LIST *to_blocks) |
| void | GetDeskewVectors (FCOORD *deskew, FCOORD *reskew) |
| void | SetEquationDetect (EquationDetectBase *detect) |
| tesseract::ColumnFinder::ColumnFinder | ( | int | gridsize, |
| const ICOORD & | bleft, | ||
| const ICOORD & | tright, | ||
| int | resolution, | ||
| TabVector_LIST * | vlines, | ||
| TabVector_LIST * | hlines, | ||
| int | vertical_x, | ||
| int | vertical_y | ||
| ) |
Definition at line 84 of file colfind.cpp.
: TabFind(gridsize, bleft, tright, vlines, vertical_x, vertical_y, resolution), min_gutter_width_(static_cast<int>(kMinGutterWidthGrid * gridsize)), mean_column_gap_(tright.x() - bleft.x()), reskew_(1.0f, 0.0f), rotation_(1.0f, 0.0f), rerotate_(1.0f, 0.0f), best_columns_(NULL), stroke_width_(NULL), part_grid_(gridsize, bleft, tright), nontext_map_(NULL), projection_(resolution), denorm_(NULL), input_blobs_win_(NULL), equation_detect_(NULL) { TabVector_IT h_it(&horizontal_lines_); h_it.add_list_after(hlines); }
| tesseract::ColumnFinder::~ColumnFinder | ( | ) | [virtual] |
Definition at line 102 of file colfind.cpp.
{
column_sets_.delete_data_pointers();
if (best_columns_ != NULL) {
delete [] best_columns_;
}
if (stroke_width_ != NULL)
delete stroke_width_;
delete input_blobs_win_;
pixDestroy(&nontext_map_);
while (denorm_ != NULL) {
DENORM* dead_denorm = denorm_;
denorm_ = const_cast<DENORM*>(denorm_->predecessor());
delete dead_denorm;
}
// The ColPartitions are destroyed automatically, but any boxes in
// the noise_parts_ list are owned and need to be deleted explicitly.
ColPartition_IT part_it(&noise_parts_);
for (part_it.mark_cycle_pt(); !part_it.cycled_list(); part_it.forward()) {
ColPartition* part = part_it.data();
part->DeleteBoxes();
}
// Likewise any boxes in the good_parts_ list need to be deleted.
// These are just the image parts. Text parts have already given their
// boxes on to the TO_BLOCK, and have empty lists.
part_it.set_to_list(&good_parts_);
for (part_it.mark_cycle_pt(); !part_it.cycled_list(); part_it.forward()) {
ColPartition* part = part_it.data();
part->DeleteBoxes();
}
// Also, any blobs on the image_bblobs_ list need to have their cblobs
// deleted. This only happens if there has been an early return from
// FindColumns, as in a normal return, the blobs go into the grid and
// end up in noise_parts_, good_parts_ or the output blocks.
BLOBNBOX_IT bb_it(&image_bblobs_);
for (bb_it.mark_cycle_pt(); !bb_it.cycled_list(); bb_it.forward()) {
BLOBNBOX* bblob = bb_it.data();
delete bblob->cblob();
}
}
| void tesseract::ColumnFinder::CorrectOrientation | ( | TO_BLOCK * | block, |
| bool | vertical_text_lines, | ||
| int | recognition_rotation | ||
| ) |
Definition at line 205 of file colfind.cpp.
{
const FCOORD anticlockwise90(0.0f, 1.0f);
const FCOORD clockwise90(0.0f, -1.0f);
const FCOORD rotation180(-1.0f, 0.0f);
const FCOORD norotation(1.0f, 0.0f);
text_rotation_ = norotation;
// Rotate the page to make the text upright, as implied by
// recognition_rotation.
rotation_ = norotation;
if (recognition_rotation == 1) {
rotation_ = anticlockwise90;
} else if (recognition_rotation == 2) {
rotation_ = rotation180;
} else if (recognition_rotation == 3) {
rotation_ = clockwise90;
}
// We infer text writing direction to be vertical if there are several
// vertical text lines detected, and horizontal if not. But if the page
// orientation was determined to be 90 or 270 degrees, the true writing
// direction is the opposite of what we inferred.
if (recognition_rotation & 1) {
vertical_text_lines = !vertical_text_lines;
}
// If we still believe the writing direction is vertical, we use the
// convention of rotating the page ccw 90 degrees to make the text lines
// horizontal, and mark the blobs for rotation cw 90 degrees for
// classification so that the text order is correct after recognition.
if (vertical_text_lines) {
rotation_.rotate(anticlockwise90);
text_rotation_.rotate(clockwise90);
}
// Set rerotate_ to the inverse of rotation_.
rerotate_ = FCOORD(rotation_.x(), -rotation_.y());
if (rotation_.x() != 1.0f || rotation_.y() != 0.0f) {
// Rotate all the blobs and tab vectors.
RotateBlobList(rotation_, &block->large_blobs);
RotateBlobList(rotation_, &block->blobs);
RotateBlobList(rotation_, &block->small_blobs);
RotateBlobList(rotation_, &block->noise_blobs);
TabFind::ResetForVerticalText(rotation_, rerotate_, &horizontal_lines_,
&min_gutter_width_);
part_grid_.Init(gridsize(), bleft(), tright());
// Reset all blobs to initial state and filter by size.
// Since they have rotated, the list they belong on could have changed.
block->ReSetAndReFilterBlobs();
SetBlockRuleEdges(block);
stroke_width_->CorrectForRotation(rerotate_, &part_grid_);
}
if (textord_debug_tabfind) {
tprintf("Vertical=%d, orientation=%d, final rotation=(%f, %f)+(%f,%f)\n",
vertical_text_lines, recognition_rotation,
rotation_.x(), rotation_.y(),
text_rotation_.x(), text_rotation_.y());
}
// Setup the denormalization.
ASSERT_HOST(denorm_ == NULL);
denorm_ = new DENORM;
denorm_->SetupNormalization(NULL, NULL, &rotation_, NULL, NULL, 0,
0.0f, 0.0f, 1.0f, 1.0f, 0.0f, 0.0f);
}
| const DENORM* tesseract::ColumnFinder::denorm | ( | ) | const [inline] |
| int tesseract::ColumnFinder::FindBlocks | ( | bool | single_column, |
| Pix * | scaled_color, | ||
| int | scaled_factor, | ||
| TO_BLOCK * | block, | ||
| Pix * | photo_mask_pix, | ||
| BLOCK_LIST * | blocks, | ||
| TO_BLOCK_LIST * | to_blocks | ||
| ) |
Definition at line 284 of file colfind.cpp.
{
pixOr(photo_mask_pix, photo_mask_pix, nontext_map_);
stroke_width_->FindLeaderPartitions(input_block, &part_grid_);
stroke_width_->RemoveLineResidue(&big_parts_);
FindInitialTabVectors(NULL, min_gutter_width_, input_block);
SetBlockRuleEdges(input_block);
stroke_width_->GradeBlobsIntoPartitions(rerotate_, input_block, nontext_map_,
denorm_, &projection_,
&part_grid_, &big_parts_);
ImageFind::FindImagePartitions(photo_mask_pix, rotation_, rerotate_,
input_block, this, &part_grid_, &big_parts_);
ImageFind::TransferImagePartsToImageMask(rerotate_, &part_grid_,
photo_mask_pix);
ImageFind::FindImagePartitions(photo_mask_pix, rotation_, rerotate_,
input_block, this, &part_grid_, &big_parts_);
part_grid_.ReTypeBlobs(&image_bblobs_);
TidyBlobs(input_block);
Reset();
// TODO(rays) need to properly handle big_parts_.
ColPartition_IT p_it(&big_parts_);
for (p_it.mark_cycle_pt(); !p_it.cycled_list(); p_it.forward())
p_it.data()->DisownBoxes();
big_parts_.clear();
delete stroke_width_;
stroke_width_ = NULL;
// A note about handling right-to-left scripts (Hebrew/Arabic):
// The columns must be reversed and come out in right-to-left instead of
// the normal left-to-right order. Because the left-to-right ordering
// is implicit in many data structures, it is simpler to fool the algorithms
// into thinking they are dealing with left-to-right text.
// To do this, we reflect the needed data in the y-axis and then reflect
// the blocks back after they have been created. This is a temporary
// arrangment that is confined to this function only, so the reflection
// is completely invisible in the output blocks.
// The only objects reflected are:
// The vertical separator lines that have already been found;
// The bounding boxes of all BLOBNBOXES on all lists on the input_block
// plus the image_bblobs. The outlines are not touched, since they are
// not looked at.
bool input_is_rtl = input_block->block->right_to_left();
if (input_is_rtl) {
// Reflect the vertical separator lines (member of TabFind).
ReflectInYAxis();
// Reflect the blob boxes.
ReflectForRtl(input_block, &image_bblobs_);
part_grid_.ReflectInYAxis();
}
if (single_column) {
// No tab stops needed. Just the grid that FindTabVectors makes.
DontFindTabVectors(&image_bblobs_, input_block, &deskew_, &reskew_);
} else {
SetBlockRuleEdges(input_block);
// Find the tab stops, estimate skew, and deskew the tabs, blobs and
// part_grid_.
FindTabVectors(&horizontal_lines_, &image_bblobs_, input_block,
min_gutter_width_, &part_grid_, &deskew_, &reskew_);
// Add the deskew to the denorm_.
DENORM* new_denorm = new DENORM;
new_denorm->SetupNormalization(NULL, NULL, &deskew_, denorm_, NULL, 0,
0.0f, 0.0f, 1.0f, 1.0f, 0.0f, 0.0f);
denorm_ = new_denorm;
}
SetBlockRuleEdges(input_block);
part_grid_.SetTabStops(this);
// Make the column_sets_.
if (!MakeColumns(single_column)) {
tprintf("Empty page!!\n");
return 0; // This is an empty page.
}
// Refill the grid using rectangular spreading, and get the benefit
// of the completed tab vectors marking the rule edges of each blob.
Clear();
#ifndef GRAPHICS_DISABLED
if (textord_tabfind_show_reject_blobs) {
ScrollView* rej_win = MakeWindow(500, 300, "Rejected blobs");
input_block->plot_graded_blobs(rej_win);
}
#endif // GRAPHICS_DISABLED
InsertBlobsToGrid(false, false, &image_bblobs_, this);
InsertBlobsToGrid(true, true, &input_block->blobs, this);
part_grid_.GridFindMargins(best_columns_);
// Split and merge the partitions by looking at local neighbours.
GridSplitPartitions();
// Resolve unknown partitions by adding to an existing partition, fixing
// the type, or declaring them noise.
part_grid_.GridFindMargins(best_columns_);
GridMergePartitions();
// Insert any unused noise blobs that are close enough to an appropriate
// partition.
InsertRemainingNoise(input_block);
// Add horizontal line separators as partitions.
GridInsertHLinePartitions();
GridInsertVLinePartitions();
// Recompute margins based on a local neighbourhood search.
part_grid_.GridFindMargins(best_columns_);
SetPartitionTypes();
if (textord_tabfind_show_initial_partitions) {
ScrollView* part_win = MakeWindow(100, 300, "InitialPartitions");
part_grid_.DisplayBoxes(part_win);
DisplayTabVectors(part_win);
}
if (equation_detect_) {
equation_detect_->FindEquationParts(&part_grid_, best_columns_);
}
if (textord_tabfind_find_tables) {
TableFinder table_finder;
table_finder.Init(gridsize(), bleft(), tright());
table_finder.set_resolution(resolution_);
table_finder.set_left_to_right_language(
!input_block->block->right_to_left());
// Copy cleaned partitions from part_grid_ to clean_part_grid_ and
// insert dot-like noise into period_grid_
table_finder.InsertCleanPartitions(&part_grid_, input_block);
// Get Table Regions
table_finder.LocateTables(&part_grid_, best_columns_, WidthCB(), reskew_);
}
GridRemoveUnderlinePartitions();
part_grid_.DeleteUnknownParts(input_block);
// Build the partitions into chains that belong in the same block and
// refine into one-to-one links, then smooth the types within each chain.
part_grid_.FindPartitionPartners();
part_grid_.FindFigureCaptions();
part_grid_.RefinePartitionPartners(true);
SmoothPartnerRuns();
#ifndef GRAPHICS_DISABLED
if (textord_tabfind_show_partitions) {
ScrollView* window = MakeWindow(400, 300, "Partitions");
if (textord_debug_images)
window->Image(AlignedBlob::textord_debug_pix().string(),
image_origin().x(), image_origin().y());
part_grid_.DisplayBoxes(window);
if (!textord_debug_printable)
DisplayTabVectors(window);
if (window != NULL && textord_tabfind_show_partitions > 1) {
delete window->AwaitEvent(SVET_DESTROY);
}
}
#endif // GRAPHICS_DISABLED
part_grid_.AssertNoDuplicates();
// Ownership of the ColPartitions moves from part_sets_ to part_grid_ here,
// and ownership of the BLOBNBOXes moves to the ColPartitions.
// (They were previously owned by the block or the image_bblobs list.)
ReleaseBlobsAndCleanupUnused(input_block);
// Ownership of the ColPartitions moves from part_grid_ to good_parts_ and
// noise_parts_ here. In text blocks, ownership of the BLOBNBOXes moves
// from the ColPartitions to the output TO_BLOCK. In non-text, the
// BLOBNBOXes stay with the ColPartitions and get deleted in the destructor.
TransformToBlocks(blocks, to_blocks);
if (textord_debug_tabfind) {
tprintf("Found %d blocks, %d to_blocks\n",
blocks->length(), to_blocks->length());
}
DisplayBlocks(blocks);
RotateAndReskewBlocks(input_is_rtl, to_blocks);
int result = 0;
#ifndef GRAPHICS_DISABLED
if (blocks_win_ != NULL) {
bool waiting = false;
do {
waiting = false;
SVEvent* event = blocks_win_->AwaitEvent(SVET_ANY);
if (event->type == SVET_INPUT && event->parameter != NULL) {
if (*event->parameter == 'd')
result = -1;
else
blocks->clear();
} else if (event->type == SVET_DESTROY) {
blocks_win_ = NULL;
} else {
waiting = true;
}
delete event;
} while (waiting);
}
#endif // GRAPHICS_DISABLED
return result;
}
Definition at line 475 of file colfind.cpp.
| bool tesseract::ColumnFinder::IsVerticallyAlignedText | ( | TO_BLOCK * | block, |
| BLOBNBOX_CLIST * | osd_blobs | ||
| ) |
Definition at line 189 of file colfind.cpp.
{
return stroke_width_->TestVerticalTextDirection(block, osd_blobs);
}
| const TextlineProjection* tesseract::ColumnFinder::projection | ( | ) | const [inline] |
| void tesseract::ColumnFinder::SetEquationDetect | ( | EquationDetectBase * | detect | ) |
Definition at line 481 of file colfind.cpp.
{
equation_detect_ = detect;
}
| void tesseract::ColumnFinder::SetupAndFilterNoise | ( | Pix * | photo_mask_pix, |
| TO_BLOCK * | input_block | ||
| ) |
Definition at line 150 of file colfind.cpp.
{
part_grid_.Init(gridsize(), bleft(), tright());
if (stroke_width_ != NULL)
delete stroke_width_;
stroke_width_ = new StrokeWidth(gridsize(), bleft(), tright());
min_gutter_width_ = static_cast<int>(kMinGutterWidthGrid * gridsize());
input_block->ReSetAndReFilterBlobs();
#ifndef GRAPHICS_DISABLED
if (textord_tabfind_show_blocks) {
input_blobs_win_ = MakeWindow(0, 0, "Filtered Input Blobs");
input_block->plot_graded_blobs(input_blobs_win_);
}
#endif // GRAPHICS_DISABLED
SetBlockRuleEdges(input_block);
pixDestroy(&nontext_map_);
// Run a preliminary strokewidth neighbour detection on the medium blobs.
stroke_width_->SetNeighboursOnMediumBlobs(input_block);
CCNonTextDetect nontext_detect(gridsize(), bleft(), tright());
// Remove obvious noise and make the initial non-text map.
nontext_map_ = nontext_detect.ComputeNonTextMask(textord_debug_tabfind,
photo_mask_pix, input_block);
// TODO(rays) experiment with making broken CJK fixing dependent on the
// language, and keeping the merged blobs on output instead of exploding at
// ColPartition::MakeBlock.
stroke_width_->FindTextlineDirectionAndFixBrokenCJK(true, input_block);
// Clear the strokewidth grid ready for rotation or leader finding.
stroke_width_->Clear();
}
1.7.5.1