TBOX PIXROW::bounding_box() const {
inT16 i;
inT16 y_coord;
inT16 min_x = MAX_INT16 - 1;
inT16 min_y = MAX_INT16 - 1;
inT16 max_x = -MAX_INT16 + 1;
inT16 max_y = -MAX_INT16 + 1;
for (i = 0; i < row_count; i++) {
y_coord = row_offset + i;
if (min[i] <= max[i]) {
if (y_coord < min_y)
min_y = y_coord;
if (y_coord + 1 > max_y)
max_y = y_coord + 1;
if (min[i] < min_x)
min_x = min[i];
if (max[i] + 1 > max_x)
max_x = max[i] + 1;
}
}
if (min_x > max_x || min_y > max_y)
return TBOX ();
else
return TBOX (ICOORD (min_x, min_y), ICOORD (max_x, max_y));
}
void BLOBNBOX::chop( //chop blobs
BLOBNBOX_IT *start_it, //location of this
BLOBNBOX_IT *end_it, //iterator
FCOORD rotation, //for landscape
float xheight //of line
) {
inT16 blobcount; //no of blobs
BLOBNBOX *newblob; //fake blob
BLOBNBOX *blob; //current blob
inT16 blobindex; //number of chop
inT16 leftx; //left edge of blob
float blobwidth; //width of each
float rightx; //right edge to scan
float ymin, ymax; //limits of new blob
float test_ymin, test_ymax; //limits of part blob
ICOORD bl, tr; //corners of box
BLOBNBOX_IT blob_it; //blob iterator
//get no of chops
blobcount = (inT16) floor (box.width () / xheight);
if (blobcount > 1 && cblob_ptr != NULL) {
//width of each
blobwidth = (float) (box.width () + 1) / blobcount;
for (blobindex = blobcount - 1, rightx = box.right ();
blobindex >= 0; blobindex--, rightx -= blobwidth) {
ymin = (float) MAX_INT32;
ymax = (float) -MAX_INT32;
blob_it = *start_it;
do {
blob = blob_it.data ();
find_cblob_vlimits(blob->cblob_ptr, rightx - blobwidth,
rightx,
/*rotation, */ test_ymin, test_ymax);
blob_it.forward ();
UpdateRange(test_ymin, test_ymax, &ymin, &ymax);
}
while (blob != end_it->data ());
if (ymin < ymax) {
leftx = (inT16) floor (rightx - blobwidth);
if (leftx < box.left ())
leftx = box.left (); //clip to real box
bl = ICOORD (leftx, (inT16) floor (ymin));
tr = ICOORD ((inT16) ceil (rightx), (inT16) ceil (ymax));
if (blobindex == 0)
box = TBOX (bl, tr); //change box
else {
newblob = new BLOBNBOX;
//box is all it has
newblob->box = TBOX (bl, tr);
//stay on current
newblob->base_char_top_ = tr.y();
newblob->base_char_bottom_ = bl.y();
end_it->add_after_stay_put (newblob);
}
}
}
}
}
// Recognizes a word or group of words, converting to WERD_RES in *words.
// Analogous to classify_word_pass1, but can handle a group of words as well.
void Tesseract::LSTMRecognizeWord(const BLOCK& block, ROW *row, WERD_RES *word,
PointerVector<WERD_RES>* words) {
TBOX word_box = word->word->bounding_box();
// Get the word image - no frills.
if (tessedit_pageseg_mode == PSM_SINGLE_WORD ||
tessedit_pageseg_mode == PSM_RAW_LINE) {
// In single word mode, use the whole image without any other row/word
// interpretation.
word_box = TBOX(0, 0, ImageWidth(), ImageHeight());
} else {
float baseline = row->base_line((word_box.left() + word_box.right()) / 2);
if (baseline + row->descenders() < word_box.bottom())
word_box.set_bottom(baseline + row->descenders());
if (baseline + row->x_height() + row->ascenders() > word_box.top())
word_box.set_top(baseline + row->x_height() + row->ascenders());
}
ImageData* im_data = GetRectImage(word_box, block, kImagePadding, &word_box);
if (im_data == NULL) return;
lstm_recognizer_->RecognizeLine(*im_data, true, classify_debug_level > 0,
kWorstDictCertainty / kCertaintyScale,
lstm_use_matrix, &unicharset, word_box, 2.0,
false, words);
delete im_data;
SearchWords(words);
}
void POLY_BLOCK::compute_bb() { //constructor
ICOORD ibl, itr; //integer bb
ICOORD botleft; //bounding box
ICOORD topright;
ICOORD pos; //current pos;
ICOORDELT_IT pts = &vertices; //iterator
botleft = *pts.data ();
topright = botleft;
do {
pos = *pts.data ();
if (pos.x () < botleft.x ())
//get bounding box
botleft = ICOORD (pos.x (), botleft.y ());
if (pos.y () < botleft.y ())
botleft = ICOORD (botleft.x (), pos.y ());
if (pos.x () > topright.x ())
topright = ICOORD (pos.x (), topright.y ());
if (pos.y () > topright.y ())
topright = ICOORD (topright.x (), pos.y ());
pts.forward ();
}
while (!pts.at_first ());
ibl = ICOORD (botleft.x (), botleft.y ());
itr = ICOORD (topright.x (), topright.y ());
box = TBOX (ibl, itr);
}
// Generates a TrainingSample from a TBLOB. Extracts features and sets
// the bounding box, so classifiers that operate on the image can work.
// TODO(rays) Make BlobToTrainingSample a member of Classify now that
// the FlexFx and FeatureDescription code have been removed and LearnBlob
// is now a member of Classify.
TrainingSample* BlobToTrainingSample(
const TBLOB& blob, bool nonlinear_norm, INT_FX_RESULT_STRUCT* fx_info,
GenericVector<INT_FEATURE_STRUCT>* bl_features) {
GenericVector<INT_FEATURE_STRUCT> cn_features;
Classify::ExtractFeatures(blob, nonlinear_norm, bl_features,
&cn_features, fx_info, nullptr);
// TODO(rays) Use blob->PreciseBoundingBox() instead.
TBOX box = blob.bounding_box();
TrainingSample* sample = nullptr;
int num_features = fx_info->NumCN;
if (num_features > 0) {
sample = TrainingSample::CopyFromFeatures(*fx_info, box, &cn_features[0],
num_features);
}
if (sample != nullptr) {
// Set the bounding box (in original image coordinates) in the sample.
TPOINT topleft, botright;
topleft.x = box.left();
topleft.y = box.top();
botright.x = box.right();
botright.y = box.bottom();
TPOINT original_topleft, original_botright;
blob.denorm().DenormTransform(nullptr, topleft, &original_topleft);
blob.denorm().DenormTransform(nullptr, botright, &original_botright);
sample->set_bounding_box(TBOX(original_topleft.x, original_botright.y,
original_botright.x, original_topleft.y));
}
return sample;
}
void OUTLINE::compute_bb() { //constructor
ICOORD ibl, itr; //integer bb
FCOORD botleft; //bounding box
FCOORD topright;
FCOORD pos; //current pos;
POLYPT_IT polypts = &outline; //iterator
botleft = polypts.data ()->pos;
topright = botleft;
start = ICOORD ((inT16) botleft.x (), (inT16) botleft.y ());
do {
pos = polypts.data ()->pos;
if (pos.x () < botleft.x ())
//get bounding box
botleft = FCOORD (pos.x (), botleft.y ());
if (pos.y () < botleft.y ())
botleft = FCOORD (botleft.x (), pos.y ());
if (pos.x () > topright.x ())
topright = FCOORD (pos.x (), topright.y ());
if (pos.y () > topright.y ())
topright = FCOORD (topright.x (), pos.y ());
polypts.forward ();
}
while (!polypts.at_first ());
ibl = ICOORD ((inT16) botleft.x (), (inT16) botleft.y ());
itr = ICOORD ((inT16) topright.x () + 1, (inT16) topright.y () + 1);
box = TBOX (ibl, itr);
}
TBOX TBOX::intersection( //shared area box
const TBOX &box) const {
inT16 left;
inT16 bottom;
inT16 right;
inT16 top;
if (overlap (box)) {
if (box.bot_left.x () > bot_left.x ())
left = box.bot_left.x ();
else
left = bot_left.x ();
if (box.top_right.x () < top_right.x ())
right = box.top_right.x ();
else
right = top_right.x ();
if (box.bot_left.y () > bot_left.y ())
bottom = box.bot_left.y ();
else
bottom = bot_left.y ();
if (box.top_right.y () < top_right.y ())
top = box.top_right.y ();
else
top = top_right.y ();
}
else {
left = MAX_INT16;
bottom = MAX_INT16;
top = -MAX_INT16;
right = -MAX_INT16;
}
return TBOX (left, bottom, right, top);
}
TBOX TBOX::bounding_union( //box enclosing both
const TBOX &box) const {
ICOORD bl; //bottom left
ICOORD tr; //top right
if (box.bot_left.x () < bot_left.x ())
bl.set_x (box.bot_left.x ());
else
bl.set_x (bot_left.x ());
if (box.top_right.x () > top_right.x ())
tr.set_x (box.top_right.x ());
else
tr.set_x (top_right.x ());
if (box.bot_left.y () < bot_left.y ())
bl.set_y (box.bot_left.y ());
else
bl.set_y (bot_left.y ());
if (box.top_right.y () > top_right.y ())
tr.set_y (box.top_right.y ());
else
tr.set_y (top_right.y ());
return TBOX (bl, tr);
}
void BLOCK::compress() { // squash it up
#define ROW_SPACING 5
ROW_IT row_it(&rows);
ROW *row;
ICOORD row_spacing (0, ROW_SPACING);
ICOORDELT_IT icoordelt_it;
sort_rows();
box = TBOX (box.topleft (), box.topleft ());
box.move_bottom_edge (ROW_SPACING);
for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
row = row_it.data ();
row->move (box.botleft () - row_spacing -
row->bounding_box ().topleft ());
box += row->bounding_box ();
}
leftside.clear ();
icoordelt_it.set_to_list (&leftside);
icoordelt_it.add_to_end (new ICOORDELT (box.left (), box.bottom ()));
icoordelt_it.add_to_end (new ICOORDELT (box.left (), box.top ()));
rightside.clear ();
icoordelt_it.set_to_list (&rightside);
icoordelt_it.add_to_end (new ICOORDELT (box.right (), box.bottom ()));
icoordelt_it.add_to_end (new ICOORDELT (box.right (), box.top ()));
}
TBOX TBOX::intersection( //shared area box
const TBOX &box) const {
ICOORD bl; //bottom left
ICOORD tr; //top right
if (overlap (box)) {
if (box.bot_left.x () > bot_left.x ())
bl.set_x (box.bot_left.x ());
else
bl.set_x (bot_left.x ());
if (box.top_right.x () < top_right.x ())
tr.set_x (box.top_right.x ());
else
tr.set_x (top_right.x ());
if (box.bot_left.y () > bot_left.y ())
bl.set_y (box.bot_left.y ());
else
bl.set_y (bot_left.y ());
if (box.top_right.y () < top_right.y ())
tr.set_y (box.top_right.y ());
else
tr.set_y (top_right.y ());
}
else {
bl.set_x (MAX_INT16);
bl.set_y (MAX_INT16);
tr.set_x (-MAX_INT16);
tr.set_y (-MAX_INT16);
}
return TBOX (bl, tr);
}
// 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;
}
TBOX M_Utils::imToCBlobGridCoords(BOX* box, PIX* im) {
inT16 height = (inT16)im->h;
inT16 left = (inT16)(box->x);
inT16 top = height - (inT16)(box->y);
inT16 right = left + (inT16)(box->w);
inT16 bottom = height - (box->y+(inT16)(box->h));
return TBOX(left,bottom,right,top);
}
// Extract the OCR results, costs (penalty points for uncertainty),
// and the bounding boxes of the characters.
static void extract_result(ELIST_ITERATOR *out,
PAGE_RES* page_res) {
PAGE_RES_IT page_res_it(page_res);
int word_count = 0;
while (page_res_it.word() != NULL) {
WERD_RES *word = page_res_it.word();
const char *str = word->best_choice->string().string();
const char *len = word->best_choice->lengths().string();
if (word_count)
add_space(out);
TBOX bln_rect;
PBLOB_LIST *blobs = word->outword->blob_list();
PBLOB_IT it(blobs);
int n = strlen(len);
TBOX** boxes_to_fix = new TBOX*[n];
for (int i = 0; i < n; i++) {
PBLOB *blob = it.data();
TBOX current = blob->bounding_box();
bln_rect = bln_rect.bounding_union(current);
TESS_CHAR *tc = new TESS_CHAR(rating_to_cost(word->best_choice->rating()),
str, *len);
tc->box = current;
boxes_to_fix[i] = &tc->box;
out->add_after_then_move(tc);
it.forward();
str += *len;
len++;
}
// Find the word bbox before normalization.
// Here we can't use the C_BLOB bboxes directly,
// since connected letters are not yet cut.
TBOX real_rect = word->word->bounding_box();
// Denormalize boxes by transforming the bbox of the whole bln word
// into the denorm bbox (`real_rect') of the whole word.
double x_stretch = double(real_rect.width()) / bln_rect.width();
double y_stretch = double(real_rect.height()) / bln_rect.height();
for (int j = 0; j < n; j++) {
TBOX *box = boxes_to_fix[j];
int x0 = int(real_rect.left() +
x_stretch * (box->left() - bln_rect.left()) + 0.5);
int x1 = int(real_rect.left() +
x_stretch * (box->right() - bln_rect.left()) + 0.5);
int y0 = int(real_rect.bottom() +
y_stretch * (box->bottom() - bln_rect.bottom()) + 0.5);
int y1 = int(real_rect.bottom() +
y_stretch * (box->top() - bln_rect.bottom()) + 0.5);
*box = TBOX(ICOORD(x0, y0), ICOORD(x1, y1));
}
delete [] boxes_to_fix;
page_res_it.forward();
word_count++;
}
}
// Parses the given box file string into a page_number, utf8_str, and
// bounding_box. Returns true on a successful parse.
// The box file is assumed to contain box definitions, one per line, of the
// following format for blob-level boxes:
// <UTF8 str> <left> <bottom> <right> <top> <page id>
// and for word/line-level boxes:
// WordStr <left> <bottom> <right> <top> <page id> #<space-delimited word str>
// See applyybox.cpp for more information.
bool ParseBoxFileStr(const char* boxfile_str, int* page_number,
STRING* utf8_str, TBOX* bounding_box) {
*bounding_box = TBOX(); // Initialize it to empty.
*utf8_str = "";
char uch[kBoxReadBufSize];
const char *buffptr = boxfile_str;
// Read the unichar without messing up on Tibetan.
// According to issue 253 the utf-8 surrogates 85 and A0 are treated
// as whitespace by sscanf, so it is more reliable to just find
// ascii space and tab.
int uch_len = 0;
// Skip unicode file designation, if present.
const unsigned char *ubuf = reinterpret_cast<const unsigned char*>(buffptr);
if (ubuf[0] == 0xef && ubuf[1] == 0xbb && ubuf[2] == 0xbf)
buffptr += 3;
// Allow a single blank as the UTF-8 string. Check for empty string and
// then blindly eat the first character.
if (*buffptr == '\0') return false;
do {
uch[uch_len++] = *buffptr++;
} while (*buffptr != '\0' && *buffptr != ' ' && *buffptr != '\t' &&
uch_len < kBoxReadBufSize - 1);
uch[uch_len] = '\0';
if (*buffptr != '\0') ++buffptr;
int x_min, y_min, x_max, y_max;
*page_number = 0;
int count = sscanf(buffptr, "%d %d %d %d %d",
&x_min, &y_min, &x_max, &y_max, page_number);
if (count != 5 && count != 4) {
tprintf("Bad box coordinates in boxfile string! %s\n", ubuf);
return false;
}
// Test for long space-delimited string label.
if (strcmp(uch, kMultiBlobLabelCode) == 0 &&
(buffptr = strchr(buffptr, '#')) != NULL) {
strncpy(uch, buffptr + 1, kBoxReadBufSize - 1);
uch[kBoxReadBufSize - 1] = '\0'; // Prevent buffer overrun.
chomp_string(uch);
uch_len = strlen(uch);
}
// Validate UTF8 by making unichars with it.
int used = 0;
while (used < uch_len) {
UNICHAR ch(uch + used, uch_len - used);
int new_used = ch.utf8_len();
if (new_used == 0) {
tprintf("Bad UTF-8 str %s starts with 0x%02x at col %d\n",
uch + used, uch[used], used + 1);
return false;
}
used += new_used;
}
*utf8_str = uch;
if (x_min > x_max) Swap(&x_min, &x_max);
if (y_min > y_max) Swap(&y_min, &y_max);
bounding_box->set_to_given_coords(x_min, y_min, x_max, y_max);
return true; // Successfully read a box.
}
// Compute the coverage and good column count. Coverage is the amount of the
// width of the page (in pixels) that is covered by ColPartitions, which are
// used to provide candidate column layouts.
// Coverage is split into good and bad. Good coverage is provided by
// ColPartitions of a frequent width (according to the callback function
// provided by TabFinder::WidthCB, which accesses stored statistics on the
// widths of ColParititions) and bad coverage is provided by all other
// ColPartitions, even if they have tab vectors at both sides. Thus:
// |-----------------------------------------------------------------|
// | Double width heading |
// |-----------------------------------------------------------------|
// |-------------------------------| |-------------------------------|
// | Common width ColParition | | Common width ColPartition |
// |-------------------------------| |-------------------------------|
// the layout with two common-width columns has better coverage than the
// double width heading, because the coverage is "good," even though less in
// total coverage than the heading, because the heading coverage is "bad."
void ColPartitionSet::ComputeCoverage() {
// Count the number of good columns and sum their width.
ColPartition_IT it(&parts_);
good_column_count_ = 0;
good_coverage_ = 0;
bad_coverage_ = 0;
bounding_box_ = TBOX();
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
ColPartition* part = it.data();
AddPartitionCoverageAndBox(*part);
}
}
/**********************************************************************
* C_OUTLINE::C_OUTLINE
*
* Constructor to build a C_OUTLINE from a C_OUTLINE_FRAG.
**********************************************************************/
C_OUTLINE::C_OUTLINE (
//constructor
//steps to copy
ICOORD startpt, DIR128 * new_steps,
inT16 length //length of loop
):start (startpt), offsets(NULL) {
inT8 dirdiff; //direction difference
DIR128 prevdir; //previous direction
DIR128 dir; //current direction
DIR128 lastdir; //dir of last step
TBOX new_box; //easy bounding
inT16 stepindex; //index to step
inT16 srcindex; //source steps
ICOORD pos; //current position
pos = startpt;
stepcount = length; // No. of steps.
ASSERT_HOST(length >= 0);
steps = reinterpret_cast<uinT8*>(alloc_mem(step_mem())); // Get memory.
memset(steps, 0, step_mem());
lastdir = new_steps[length - 1];
prevdir = lastdir;
for (stepindex = 0, srcindex = 0; srcindex < length;
stepindex++, srcindex++) {
new_box = TBOX (pos, pos);
box += new_box;
//copy steps
dir = new_steps[srcindex];
set_step(stepindex, dir);
dirdiff = dir - prevdir;
pos += step (stepindex);
if ((dirdiff == 64 || dirdiff == -64) && stepindex > 0) {
stepindex -= 2; //cancel there-and-back
prevdir = stepindex >= 0 ? step_dir (stepindex) : lastdir;
}
else
prevdir = dir;
}
ASSERT_HOST (pos.x () == startpt.x () && pos.y () == startpt.y ());
do {
dirdiff = step_dir (stepindex - 1) - step_dir (0);
if (dirdiff == 64 || dirdiff == -64) {
start += step (0);
stepindex -= 2; //cancel there-and-back
for (int i = 0; i < stepindex; ++i)
set_step(i, step_dir(i + 1));
}
}
while (stepindex > 1 && (dirdiff == 64 || dirdiff == -64));
stepcount = stepindex;
ASSERT_HOST (stepcount >= 4);
}
// Return the bounding boxes of columns at the given y-range
void ColPartitionSet::GetColumnBoxes(int y_bottom, int y_top,
ColSegment_LIST *segments) {
ColPartition_IT it(&parts_);
ColSegment_IT col_it(segments);
col_it.move_to_last();
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
ColPartition* part = it.data();
ICOORD bot_left(part->LeftAtY(y_top), y_bottom);
ICOORD top_right(part->RightAtY(y_bottom), y_top);
ColSegment *col_seg = new ColSegment();
col_seg->InsertBox(TBOX(bot_left, top_right));
col_it.add_after_then_move(col_seg);
}
}
/**********************************************************************
* char_box_to_tbox
*
* Create a TBOX from a character bounding box. If nonzero, the
* x_offset accounts for any additional padding of the word box that
* should be taken into account.
*
**********************************************************************/
TBOX char_box_to_tbox(Box* char_box, TBOX word_box, int x_offset) {
l_int32 left;
l_int32 top;
l_int32 width;
l_int32 height;
l_int32 right;
l_int32 bottom;
boxGetGeometry(char_box, &left, &top, &width, &height);
left += word_box.left() - x_offset;
right = left + width;
top = word_box.bottom() + word_box.height() - top;
bottom = top - height;
return TBOX(left, bottom, right, top);
}
// Parses the given box file string into a page_number, utf8_str, and
// bounding_box. Returns true on a successful parse.
// The box file is assumed to contain box definitions, one per line, of the
// following format for blob-level boxes:
// <UTF8 str> <left> <bottom> <right> <top> <page id>
// and for word/line-level boxes:
// WordStr <left> <bottom> <right> <top> <page id> #<space-delimited word str>
// See applyybox.cpp for more information.
bool ParseBoxFileStr(const char* boxfile_str, int* page_number,
STRING* utf8_str, TBOX* bounding_box) {
*bounding_box = TBOX(); // Initialize it to empty.
*utf8_str = "";
char uch[kBoxReadBufSize];
const char *buffptr = boxfile_str;
// Read the unichar without messing up on Tibetan.
// According to issue 253 the utf-8 surrogates 85 and A0 are treated
// as whitespace by sscanf, so it is more reliable to just find
// ascii space and tab.
int uch_len = 0;
while (*buffptr != '\0' && *buffptr != ' ' && *buffptr != '\t' &&
uch_len < kBoxReadBufSize - 1) {
uch[uch_len++] = *buffptr++;
}
uch[uch_len] = '\0';
if (*buffptr != '\0') ++buffptr;
int x_min, y_min, x_max, y_max;
*page_number = 0;
int count = sscanf(buffptr, "%d %d %d %d %d",
&x_min, &y_min, &x_max, &y_max, page_number);
if (count != 5 && count != 4) {
tprintf("Bad box coordinates in boxfile string!\n");
return false;
}
// Test for long space-delimited string label.
if (strcmp(uch, kMultiBlobLabelCode) == 0 &&
(buffptr = strchr(buffptr, '#')) != NULL) {
strncpy(uch, buffptr + 1, kBoxReadBufSize);
chomp_string(uch);
uch_len = strlen(uch);
}
// Validate UTF8 by making unichars with it.
int used = 0;
while (used < uch_len) {
UNICHAR ch(uch + used, uch_len - used);
int new_used = ch.utf8_len();
if (new_used == 0) {
tprintf("Bad UTF-8 str %s starts with 0x%02x at col %d\n",
uch + used, uch[used], used + 1);
return false;
}
used += new_used;
}
*utf8_str = uch;
bounding_box->set_to_given_coords(x_min, y_min, x_max, y_max);
return true; // Successfully read a box.
}
ScrollView* display_clip_image(WERD *word, //word to be processed
IMAGE &bin_image, //whole image
PIXROW_LIST *pixrow_list, //pixrows built
TBOX &pix_box //box of subimage
) {
ScrollView* clip_window; //window for debug
TBOX word_box = word->bounding_box ();
int border = word_box.height () / 2;
TBOX display_box = word_box;
display_box.move_bottom_edge (-border);
display_box.move_top_edge (border);
display_box.move_left_edge (-border);
display_box.move_right_edge (border);
display_box -= TBOX (ICOORD (0, 0 - BUG_OFFSET),
ICOORD (bin_image.get_xsize (),
bin_image.get_ysize () - BUG_OFFSET));
pgeditor_msg ("Creating Clip window...");
clip_window = new ScrollView("Clipped Blobs",
editor_word_xpos, editor_word_ypos,
3 * (word_box.width () + 2 * border),
3 * (word_box.height () + 2 * border),
display_box.left () + display_box.right (),
display_box.bottom () - BUG_OFFSET +
display_box.top () - BUG_OFFSET,
true);
// ymin, ymax
pgeditor_msg ("Creating Clip window...Done");
clip_window->Clear();
sv_show_sub_image (&bin_image,
display_box.left (),
display_box.bottom (),
display_box.width (),
display_box.height (),
clip_window,
display_box.left (), display_box.bottom () - BUG_OFFSET);
word->plot (clip_window, ScrollView::RED);
word_box.plot (clip_window, ScrollView::BLUE, ScrollView::BLUE);
pix_box.plot (clip_window, ScrollView::BLUE, ScrollView::BLUE);
plot_pixrows(pixrow_list, clip_window);
return clip_window;
}
SEAM *Wordrec::chop_overlapping_blob(const GenericVector<TBOX>& boxes,
bool italic_blob, WERD_RES *word_res,
int *blob_number) {
TWERD *word = word_res->chopped_word;
for (*blob_number = 0; *blob_number < word->NumBlobs(); ++*blob_number) {
TBLOB *blob = word->blobs[*blob_number];
TPOINT topleft, botright;
topleft.x = blob->bounding_box().left();
topleft.y = blob->bounding_box().top();
botright.x = blob->bounding_box().right();
botright.y = blob->bounding_box().bottom();
TPOINT original_topleft, original_botright;
word_res->denorm.DenormTransform(NULL, topleft, &original_topleft);
word_res->denorm.DenormTransform(NULL, botright, &original_botright);
TBOX original_box = TBOX(original_topleft.x, original_botright.y,
original_botright.x, original_topleft.y);
bool almost_equal_box = false;
int num_overlap = 0;
for (int i = 0; i < boxes.size(); i++) {
if (original_box.overlap_fraction(boxes[i]) > 0.125)
num_overlap++;
if (original_box.almost_equal(boxes[i], 3))
almost_equal_box = true;
}
TPOINT location;
if (divisible_blob(blob, italic_blob, &location) ||
(!almost_equal_box && num_overlap > 1)) {
SEAM *seam = attempt_blob_chop(word, blob, *blob_number,
italic_blob, word_res->seam_array);
if (seam != NULL)
return seam;
}
}
*blob_number = -1;
return NULL;
}
// Returns the bounding box of all the points in the split.
TBOX SPLIT::bounding_box() const {
return TBOX(
MIN(point1->pos.x, point2->pos.x), MIN(point1->pos.y, point2->pos.y),
MAX(point1->pos.x, point2->pos.x), MAX(point1->pos.y, point2->pos.y));
}
/**
* process_image_event()
*
* User has done something in the image window - mouse down or up. Work out
* what it is and do something with it.
* If DOWN - just remember where it was.
* If UP - for each word in the selected area do the operation defined by
* the current mode.
*/
void Tesseract::process_image_event( // action in image win
const SVEvent &event) {
// The following variable should remain static, since it is used by
// debug editor, which uses a single Tesseract instance.
static ICOORD down;
ICOORD up;
TBOX selection_box;
char msg[80];
switch(event.type) {
case SVET_SELECTION:
if (event.type == SVET_SELECTION) {
down.set_x(event.x + event.x_size);
down.set_y(event.y + event.y_size);
if (mode == SHOW_POINT_CMD_EVENT)
show_point(current_page_res, event.x, event.y);
}
up.set_x(event.x);
up.set_y(event.y);
selection_box = TBOX(down, up);
switch(mode) {
case CHANGE_DISP_CMD_EVENT:
process_selected_words(
current_page_res,
selection_box,
&tesseract::Tesseract::word_blank_and_set_display);
break;
case DUMP_WERD_CMD_EVENT:
process_selected_words(current_page_res,
selection_box,
&tesseract::Tesseract::word_dumper);
break;
case SHOW_BLN_WERD_CMD_EVENT:
process_selected_words(current_page_res,
selection_box,
&tesseract::Tesseract::word_bln_display);
break;
case DEBUG_WERD_CMD_EVENT:
debug_word(current_page_res, selection_box);
break;
case SHOW_POINT_CMD_EVENT:
break; // ignore up event
case RECOG_WERDS:
image_win->AddMessage("Recogging selected words");
this->process_selected_words(current_page_res,
selection_box,
&Tesseract::recog_interactive);
break;
case RECOG_PSEUDO:
image_win->AddMessage("Recogging selected blobs");
recog_pseudo_word(current_page_res, selection_box);
break;
default:
sprintf(msg, "Mode %d not yet implemented", mode);
image_win->AddMessage(msg);
break;
}
default:
break;
}
}
void char_clip_word( //
WERD *word, //word to be processed
IMAGE &bin_image, //whole image
PIXROW_LIST *&pixrow_list, //pixrows built
IMAGELINE *&imlines, //lines cut from image
TBOX &pix_box //box defining imlines
) {
TBOX word_box = word->bounding_box ();
PBLOB_LIST *blob_list;
PBLOB_IT blob_it;
PIXROW_IT pixrow_it;
inT16 pix_offset; //Y pos of pixrow[0]
inT16 row_height; //No of pix rows
inT16 imlines_x_offset;
PIXROW *prev;
PIXROW *next;
PIXROW *current;
BOOL8 changed; //still improving
BOOL8 just_changed; //still improving
inT16 iteration_count = 0;
inT16 foreground_colour;
if (word->flag (W_INVERSE))
foreground_colour = 1;
else
foreground_colour = 0;
/* Define region for max pixrow expansion */
pix_box = word_box;
pix_box.move_bottom_edge (-pix_word_margin);
pix_box.move_top_edge (pix_word_margin);
pix_box.move_left_edge (-pix_word_margin);
pix_box.move_right_edge (pix_word_margin);
pix_box -= TBOX (ICOORD (0, 0 + BUG_OFFSET),
ICOORD (bin_image.get_xsize (),
bin_image.get_ysize () - BUG_OFFSET));
/* Generate pixrows list */
pix_offset = pix_box.bottom ();
row_height = pix_box.height ();
blob_list = word->blob_list ();
blob_it.set_to_list (blob_list);
pixrow_list = new PIXROW_LIST;
pixrow_it.set_to_list (pixrow_list);
for (blob_it.mark_cycle_pt (); !blob_it.cycled_list (); blob_it.forward ()) {
PIXROW *row = new PIXROW (pix_offset, row_height, blob_it.data ());
ASSERT_HOST (!row->
bad_box (bin_image.get_xsize (), bin_image.get_ysize ()));
pixrow_it.add_after_then_move (row);
}
imlines = generate_imlines (bin_image, pix_box);
/* Contract pixrows - shrink min and max back to black pixels */
imlines_x_offset = pix_box.left ();
pixrow_it.move_to_first ();
for (pixrow_it.mark_cycle_pt ();
!pixrow_it.cycled_list (); pixrow_it.forward ()) {
ASSERT_HOST (!pixrow_it.data ()->
bad_box (bin_image.get_xsize (), bin_image.get_ysize ()));
pixrow_it.data ()->contract (imlines, imlines_x_offset,
foreground_colour);
ASSERT_HOST (!pixrow_it.data ()->
bad_box (bin_image.get_xsize (), bin_image.get_ysize ()));
}
/* Expand pixrows iteratively 1 pixel at a time */
do {
changed = FALSE;
pixrow_it.move_to_first ();
prev = NULL;
current = NULL;
next = pixrow_it.data ();
for (pixrow_it.mark_cycle_pt ();
!pixrow_it.cycled_list (); pixrow_it.forward ()) {
prev = current;
current = next;
if (pixrow_it.at_last ())
next = NULL;
else
next = pixrow_it.data_relative (1);
just_changed = current->extend (imlines, pix_box, prev, next,
foreground_colour);
ASSERT_HOST (!current->
bad_box (bin_image.get_xsize (),
bin_image.get_ysize ()));
changed = changed || just_changed;
}
iteration_count++;
}
while (changed);
}
/// Gather consecutive blobs that match the given box into the best_state
/// and corresponding correct_text.
///
/// Fights over which box owns which blobs are settled by pre-chopping and
/// 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 appropriate blob for a box.
///
/// This means that occasionally, blobs may be incorrectly segmented if the
/// chopper fails to find a suitable chop point.
bool Tesseract::ResegmentCharBox(PAGE_RES* page_res, const TBOX *prev_box,
const TBOX& box, const TBOX& next_box,
const char* correct_text) {
if (applybox_debug > 1) {
tprintf("\nAPPLY_BOX: in ResegmentCharBox() for %s\n", correct_text);
}
PAGE_RES_IT page_res_it(page_res);
WERD_RES* word_res;
for (word_res = page_res_it.word(); word_res != NULL;
word_res = page_res_it.forward()) {
if (!word_res->box_word->bounding_box().major_overlap(box))
continue;
if (applybox_debug > 1) {
tprintf("Checking word box:");
word_res->box_word->bounding_box().print();
}
int word_len = word_res->box_word->length();
for (int i = 0; i < word_len; ++i) {
TBOX char_box = TBOX();
int blob_count = 0;
for (blob_count = 0; i + blob_count < word_len; ++blob_count) {
TBOX blob_box = word_res->box_word->BlobBox(i + blob_count);
if (!blob_box.major_overlap(box))
break;
if (word_res->correct_text[i + blob_count].length() > 0)
break; // Blob is claimed already.
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)
break; // Blob is a better match for next box.
char_box += blob_box;
}
if (blob_count > 0) {
if (applybox_debug > 1) {
tprintf("Index [%d, %d) seem good.\n", i, i + blob_count);
}
if (!char_box.almost_equal(box, 3) &&
(box.x_gap(next_box) < -3 ||
(prev_box != NULL && prev_box->x_gap(box) < -3))) {
return false;
}
// We refine just the box_word, best_state and correct_text here.
// The rebuild_word is made in TidyUp.
// blob_count blobs are put together to match the box. Merge the
// box_word boxes, save the blob_count in the state and the text.
word_res->box_word->MergeBoxes(i, i + blob_count);
word_res->best_state[i] = blob_count;
word_res->correct_text[i] = correct_text;
if (applybox_debug > 2) {
tprintf("%d Blobs match: blob box:", blob_count);
word_res->box_word->BlobBox(i).print();
tprintf("Matches box:");
box.print();
tprintf("With next box:");
next_box.print();
}
// Eliminated best_state and correct_text entries for the consumed
// blobs.
for (int j = 1; j < blob_count; ++j) {
word_res->best_state.remove(i + 1);
word_res->correct_text.remove(i + 1);
}
// Assume that no box spans multiple source words, so we are done with
// this box.
if (applybox_debug > 1) {
tprintf("Best state = ");
for (int j = 0; j < word_res->best_state.size(); ++j) {
tprintf("%d ", word_res->best_state[j]);
}
tprintf("\n");
tprintf("Correct text = [[ ");
for (int j = 0; j < word_res->correct_text.size(); ++j) {
tprintf("%s ", word_res->correct_text[j].string());
}
tprintf("]]\n");
}
return true;
}
}
}
if (applybox_debug > 0) {
tprintf("FAIL!\n");
}
return false; // Failure.
}
C_OUTLINE::C_OUTLINE( //constructor
C_OUTLINE *srcline, //outline to
FCOORD rotation //rotate
) : offsets(NULL) {
TBOX new_box; //easy bounding
inT16 stepindex; //index to step
inT16 dirdiff; //direction change
ICOORD pos; //current position
ICOORD prevpos; //previous dest point
ICOORD destpos; //destination point
inT16 destindex; //index to step
DIR128 dir; //coded direction
uinT8 new_step;
stepcount = srcline->stepcount * 2;
if (stepcount == 0) {
steps = NULL;
box = srcline->box;
box.rotate(rotation);
return;
}
//get memory
steps = (uinT8 *) alloc_mem (step_mem());
memset(steps, 0, step_mem());
for (int iteration = 0; iteration < 2; ++iteration) {
DIR128 round1 = iteration == 0 ? 32 : 0;
DIR128 round2 = iteration != 0 ? 32 : 0;
pos = srcline->start;
prevpos = pos;
prevpos.rotate (rotation);
start = prevpos;
box = TBOX (start, start);
destindex = 0;
for (stepindex = 0; stepindex < srcline->stepcount; stepindex++) {
pos += srcline->step (stepindex);
destpos = pos;
destpos.rotate (rotation);
// tprintf("%i %i %i %i ", destpos.x(), destpos.y(), pos.x(), pos.y());
while (destpos.x () != prevpos.x () || destpos.y () != prevpos.y ()) {
dir = DIR128 (FCOORD (destpos - prevpos));
dir += 64; //turn to step style
new_step = dir.get_dir ();
// tprintf(" %i\n", new_step);
if (new_step & 31) {
set_step(destindex++, dir + round1);
prevpos += step(destindex - 1);
if (destindex < 2
|| ((dirdiff =
step_dir (destindex - 1) - step_dir (destindex - 2)) !=
-64 && dirdiff != 64)) {
set_step(destindex++, dir + round2);
prevpos += step(destindex - 1);
} else {
prevpos -= step(destindex - 1);
destindex--;
prevpos -= step(destindex - 1);
set_step(destindex - 1, dir + round2);
prevpos += step(destindex - 1);
}
}
else {
set_step(destindex++, dir);
prevpos += step(destindex - 1);
}
while (destindex >= 2 &&
((dirdiff =
step_dir (destindex - 1) - step_dir (destindex - 2)) == -64 ||
dirdiff == 64)) {
prevpos -= step(destindex - 1);
prevpos -= step(destindex - 2);
destindex -= 2; // Forget u turn
}
//ASSERT_HOST(prevpos.x() == destpos.x() && prevpos.y() == destpos.y());
new_box = TBOX (destpos, destpos);
box += new_box;
}
}
ASSERT_HOST (destpos.x () == start.x () && destpos.y () == start.y ());
dirdiff = step_dir (destindex - 1) - step_dir (0);
while ((dirdiff == 64 || dirdiff == -64) && destindex > 1) {
start += step (0);
destindex -= 2;
for (int i = 0; i < destindex; ++i)
set_step(i, step_dir(i + 1));
dirdiff = step_dir (destindex - 1) - step_dir (0);
}
if (destindex >= 4)
break;
}
ASSERT_HOST(destindex <= stepcount);
stepcount = destindex;
destpos = start;
for (stepindex = 0; stepindex < stepcount; stepindex++) {
destpos += step (stepindex);
}
ASSERT_HOST (destpos.x () == start.x () && destpos.y () == start.y ());
}
TBOX TESSLINE::bounding_box() const {
return TBOX(topleft.x, botright.y, botright.x, topleft.y);
}
