/**********************************************************************
* run_cube_combiner
*
* Iterates through tesseract's results and calls cube on each word,
* combining the results with the existing tesseract result.
**********************************************************************/
void Tesseract::run_cube_combiner(PAGE_RES *page_res) {
if (page_res == NULL || tess_cube_combiner_ == NULL)
return;
PAGE_RES_IT page_res_it(page_res);
// Iterate through the word results and call cube on each word.
for (page_res_it.restart_page(); page_res_it.word () != NULL;
page_res_it.forward()) {
BLOCK* block = page_res_it.block()->block;
if (block->poly_block() != NULL && !block->poly_block()->IsText())
continue; // Don't deal with non-text blocks.
WERD_RES* word = page_res_it.word();
// Skip cube entirely if tesseract's certainty is greater than threshold.
int combiner_run_thresh = convert_prob_to_tess_certainty(
cube_cntxt_->Params()->CombinerRunThresh());
if (word->best_choice->certainty() >= combiner_run_thresh) {
continue;
}
// Use the same language as Tesseract used for the word.
Tesseract* lang_tess = word->tesseract;
// Setup a trial WERD_RES in which to classify with cube.
WERD_RES cube_word;
cube_word.InitForRetryRecognition(*word);
cube_word.SetupForRecognition(lang_tess->unicharset, this, BestPix(),
OEM_CUBE_ONLY,
NULL, false, false, false,
page_res_it.row()->row,
page_res_it.block()->block);
CubeObject *cube_obj = lang_tess->cube_recognize_word(
page_res_it.block()->block, &cube_word);
if (cube_obj != NULL)
lang_tess->cube_combine_word(cube_obj, &cube_word, word);
delete cube_obj;
}
}
/// Tests the chopper by exhaustively running chop_one_blob.
/// The word_res will contain filled chopped_word, seam_array, denorm,
/// box_word and best_state for the maximally chopped word.
void Tesseract::MaximallyChopWord(const GenericVector<TBOX>& boxes,
BLOCK* block, ROW* row,
WERD_RES* word_res) {
if (!word_res->SetupForRecognition(unicharset, this, BestPix(),
tessedit_ocr_engine_mode, NULL,
classify_bln_numeric_mode,
textord_use_cjk_fp_model,
poly_allow_detailed_fx,
row, block)) {
word_res->CloneChoppedToRebuild();
return;
}
if (chop_debug) {
tprintf("Maximally chopping word at:");
word_res->word->bounding_box().print();
}
GenericVector<BLOB_CHOICE*> blob_choices;
ASSERT_HOST(!word_res->chopped_word->blobs.empty());
float rating = static_cast<float>(MAX_INT8);
for (int i = 0; i < word_res->chopped_word->NumBlobs(); ++i) {
// The rating and certainty are not quite arbitrary. Since
// select_blob_to_chop uses the worst certainty to choose, they all have
// to be different, so starting with MAX_INT8, subtract 1/8 for each blob
// in here, and then divide by e each time they are chopped, which
// should guarantee a set of unequal values for the whole tree of blobs
// produced, however much chopping is required. The chops are thus only
// limited by the ability of the chopper to find suitable chop points,
// and not by the value of the certainties.
BLOB_CHOICE* choice =
new BLOB_CHOICE(0, rating, -rating, -1, 0.0f, 0.0f, 0.0f, BCC_FAKE);
blob_choices.push_back(choice);
rating -= 0.125f;
}
const double e = exp(1.0); // The base of natural logs.
int blob_number;
int right_chop_index = 0;
if (!assume_fixed_pitch_char_segment) {
// We only chop if the language is not fixed pitch like CJK.
SEAM* seam = NULL;
while ((seam = chop_one_blob(boxes, blob_choices, word_res,
&blob_number)) != NULL) {
word_res->InsertSeam(blob_number, seam);
BLOB_CHOICE* left_choice = blob_choices[blob_number];
rating = left_choice->rating() / e;
left_choice->set_rating(rating);
left_choice->set_certainty(-rating);
// combine confidence w/ serial #
BLOB_CHOICE* right_choice = new BLOB_CHOICE(++right_chop_index,
rating - 0.125f, -rating, -1,
0.0f, 0.0f, 0.0f, BCC_FAKE);
blob_choices.insert(right_choice, blob_number + 1);
}
}
word_res->CloneChoppedToRebuild();
word_res->FakeClassifyWord(blob_choices.size(), &blob_choices[0]);
}
// Helper gets the image of a rectangle, using the block.re_rotation() if
// needed to get to the image, and rotating the result back to horizontal
// layout. (CJK characters will be on their left sides) The vertical text flag
// is set in the returned ImageData if the text was originally vertical, which
// can be used to invoke a different CJK recognition engine. The revised_box
// is also returned to enable calculation of output bounding boxes.
ImageData* Tesseract::GetRectImage(const TBOX& box, const BLOCK& block,
int padding, TBOX* revised_box) const {
TBOX wbox = box;
wbox.pad(padding, padding);
*revised_box = wbox;
// Number of clockwise 90 degree rotations needed to get back to tesseract
// coords from the clipped image.
int num_rotations = 0;
if (block.re_rotation().y() > 0.0f)
num_rotations = 1;
else if (block.re_rotation().x() < 0.0f)
num_rotations = 2;
else if (block.re_rotation().y() < 0.0f)
num_rotations = 3;
// Handle two cases automatically: 1 the box came from the block, 2 the box
// came from a box file, and refers to the image, which the block may not.
if (block.bounding_box().major_overlap(*revised_box))
revised_box->rotate(block.re_rotation());
// Now revised_box always refers to the image.
// BestPix is never colormapped, but may be of any depth.
Pix* pix = BestPix();
int width = pixGetWidth(pix);
int height = pixGetHeight(pix);
TBOX image_box(0, 0, width, height);
// Clip to image bounds;
*revised_box &= image_box;
if (revised_box->null_box()) return NULL;
Box* clip_box = boxCreate(revised_box->left(), height - revised_box->top(),
revised_box->width(), revised_box->height());
Pix* box_pix = pixClipRectangle(pix, clip_box, NULL);
if (box_pix == NULL) return NULL;
boxDestroy(&clip_box);
if (num_rotations > 0) {
Pix* rot_pix = pixRotateOrth(box_pix, num_rotations);
pixDestroy(&box_pix);
box_pix = rot_pix;
}
// Convert sub-8-bit images to 8 bit.
int depth = pixGetDepth(box_pix);
if (depth < 8) {
Pix* grey;
grey = pixConvertTo8(box_pix, false);
pixDestroy(&box_pix);
box_pix = grey;
}
bool vertical_text = false;
if (num_rotations > 0) {
// Rotated the clipped revised box back to internal coordinates.
FCOORD rotation(block.re_rotation().x(), -block.re_rotation().y());
revised_box->rotate(rotation);
if (num_rotations != 2)
vertical_text = true;
}
return new ImageData(vertical_text, box_pix);
}
/**
* word_bln_display()
*
* Normalize word and display in word window
*/
BOOL8 Tesseract::word_bln_display(BLOCK* block, ROW* row, WERD_RES* word_res) {
TWERD *bln_word = word_res->chopped_word;
if (bln_word == NULL) {
word_res->SetupForTessRecognition(unicharset, this, BestPix(), false,
this->textord_use_cjk_fp_model,
row, block);
bln_word = word_res->chopped_word;
}
bln_word_window_handle()->Clear();
display_bln_lines(bln_word_window_handle(), ScrollView::CYAN,
1.0, 0.0f, -1000.0f, 1000.0f);
bln_word->plot(bln_word_window_handle());
bln_word_window_handle()->Update();
return TRUE;
}
// page_res is non-const because the iterator doesn't know if you are going
// to change the items it points to! Really a const here though.
void Tesseract::blob_feature_display(PAGE_RES* page_res,
const TBOX& selection_box) {
ROW* row; // row of word
BLOCK* block; // block of word
WERD* word = make_pseudo_word(page_res, selection_box, block, row);
if (word != NULL) {
WERD_RES word_res(word);
word_res.x_height = row->x_height();
word_res.SetupForRecognition(unicharset, this, BestPix(),
tessedit_ocr_engine_mode, NULL,
classify_bln_numeric_mode,
textord_use_cjk_fp_model,
poly_allow_detailed_fx,
row, block);
TWERD* bln_word = word_res.chopped_word;
TBLOB* bln_blob = bln_word->blobs[0];
INT_FX_RESULT_STRUCT fx_info;
GenericVector<INT_FEATURE_STRUCT> bl_features;
GenericVector<INT_FEATURE_STRUCT> cn_features;
Classify::ExtractFeatures(*bln_blob, classify_nonlinear_norm, &bl_features,
&cn_features, &fx_info, NULL);
// Display baseline features.
ScrollView* bl_win = CreateFeatureSpaceWindow("BL Features", 512, 0);
ClearFeatureSpaceWindow(baseline, bl_win);
for (int f = 0; f < bl_features.size(); ++f)
RenderIntFeature(bl_win, &bl_features[f], ScrollView::GREEN);
bl_win->Update();
// Display cn features.
ScrollView* cn_win = CreateFeatureSpaceWindow("CN Features", 512, 0);
ClearFeatureSpaceWindow(character, cn_win);
for (int f = 0; f < cn_features.size(); ++f)
RenderIntFeature(cn_win, &cn_features[f], ScrollView::GREEN);
cn_win->Update();
delete word;
}
}
/**
* word_bln_display()
*
* Normalize word and display in word window
*/
BOOL8 Tesseract::word_bln_display(PAGE_RES_IT* pr_it) {
WERD_RES* word_res = pr_it->word();
if (word_res->chopped_word == NULL) {
// Setup word normalization parameters.
word_res->SetupForRecognition(unicharset, this, BestPix(),
tessedit_ocr_engine_mode, NULL,
classify_bln_numeric_mode,
textord_use_cjk_fp_model,
poly_allow_detailed_fx,
pr_it->row()->row, pr_it->block()->block);
}
bln_word_window_handle()->Clear();
display_bln_lines(bln_word_window_handle(), ScrollView::CYAN,
1.0, 0.0f, -1000.0f, 1000.0f);
C_BLOB_IT it(word_res->word->cblob_list());
ScrollView::Color color = WERD::NextColor(ScrollView::BLACK);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
it.data()->plot_normed(word_res->denorm, color, ScrollView::BROWN,
bln_word_window_handle());
color = WERD::NextColor(color);
}
bln_word_window_handle()->Update();
return TRUE;
}
