int main(int argc, char** argv) {
int option;
const char* output_directory = ".";
STRING unicharset_file_name;
// Special characters are now included by default.
UNICHARSET unicharset;
setlocale(LC_ALL, "");
// Print usage
if (argc <= 1) {
printf("Usage: %s [-D DIRECTORY] FILE...\n", argv[0]);
exit(1);
}
// Parse arguments
while ((option = tessopt(argc, argv, "D" )) != EOF) {
switch (option) {
case 'D':
output_directory = tessoptarg;
++tessoptind;
break;
}
}
// Save file name
unicharset_file_name = output_directory;
unicharset_file_name += "/";
unicharset_file_name += kUnicharsetFileName;
// Load box files
for (; tessoptind < argc; ++tessoptind) {
printf("Extracting unicharset from %s\n", argv[tessoptind]);
FILE* box_file = fopen(argv[tessoptind], "rb");
if (box_file == NULL) {
printf("Cannot open box file %s\n", argv[tessoptind]);
return -1;
}
TBOX box;
STRING unichar_string;
int line_number = 0;
while (ReadNextBox(&line_number, box_file, &unichar_string, &box)) {
unicharset.unichar_insert(unichar_string.string());
set_properties(&unicharset, unichar_string.string());
}
}
// Write unicharset file
if (unicharset.save_to_file(unicharset_file_name.string())) {
printf("Wrote unicharset file %s.\n", unicharset_file_name.string());
}
else {
printf("Cannot save unicharset file %s.\n", unicharset_file_name.string());
return -1;
}
return 0;
}
// This function takes tif/box pair of files and runs recognition on the image,
// while making sure that the word bounds that tesseract identified roughly
// match to those specified by the input box file. For each word (ngram in a
// single bounding box from the input box file) it outputs the ocred result,
// the correct label, rating and certainty.
void Tesseract::recog_training_segmented(const STRING &fname,
PAGE_RES *page_res,
volatile ETEXT_DESC *monitor,
FILE *output_file) {
STRING box_fname = fname;
const char *lastdot = strrchr(box_fname.string(), '.');
if (lastdot != NULL) box_fname[lastdot - box_fname.string()] = '\0';
box_fname += ".box";
// ReadNextBox() will close box_file
FILE *box_file = open_file(box_fname.string(), "r");
PAGE_RES_IT page_res_it;
page_res_it.page_res = page_res;
page_res_it.restart_page();
STRING label;
// Process all the words on this page.
TBOX tbox; // tesseract-identified box
TBOX bbox; // box from the box file
bool keep_going;
int line_number = 0;
int examined_words = 0;
do {
keep_going = read_t(&page_res_it, &tbox);
keep_going &= ReadNextBox(applybox_page, &line_number, box_file, &label,
&bbox);
// Align bottom left points of the TBOXes.
while (keep_going &&
!NearlyEqual<int>(tbox.bottom(), bbox.bottom(), kMaxBoxEdgeDiff)) {
if (bbox.bottom() < tbox.bottom()) {
page_res_it.forward();
keep_going = read_t(&page_res_it, &tbox);
} else {
keep_going = ReadNextBox(applybox_page, &line_number, box_file, &label,
&bbox);
}
}
while (keep_going &&
!NearlyEqual<int>(tbox.left(), bbox.left(), kMaxBoxEdgeDiff)) {
if (bbox.left() > tbox.left()) {
page_res_it.forward();
keep_going = read_t(&page_res_it, &tbox);
} else {
keep_going = ReadNextBox(applybox_page, &line_number, box_file, &label,
&bbox);
}
}
// OCR the word if top right points of the TBOXes are similar.
if (keep_going &&
NearlyEqual<int>(tbox.right(), bbox.right(), kMaxBoxEdgeDiff) &&
NearlyEqual<int>(tbox.top(), bbox.top(), kMaxBoxEdgeDiff)) {
ambigs_classify_and_output(label.string(), &page_res_it, output_file);
examined_words++;
}
page_res_it.forward();
} while (keep_going);
// Set up scripts on all of the words that did not get sent to
// ambigs_classify_and_output. They all should have, but if all the
// werd_res's don't get uch_sets, tesseract will crash when you try
// to iterate over them. :-(
int total_words = 0;
for (page_res_it.restart_page(); page_res_it.block() != NULL;
page_res_it.forward()) {
if (page_res_it.word()) {
if (page_res_it.word()->uch_set == NULL)
page_res_it.word()->SetupFake(unicharset);
total_words++;
}
}
if (examined_words < 0.85 * total_words) {
tprintf("TODO(antonova): clean up recog_training_segmented; "
" It examined only a small fraction of the ambigs image.\n");
}
tprintf("recog_training_segmented: examined %d / %d words.\n",
examined_words, total_words);
}
// TODO(rays) convert all uses of ReadNextBox to use the new ReadAllBoxes.
// Box files are used ONLY DURING TRAINING, but by both processes of
// creating tr files with tesseract, and unicharset_extractor.
// ReadNextBox factors out the code to interpret a line of a box
// file so that applybox and unicharset_extractor interpret the same way.
// This function returns the next valid box file utf8 string and coords
// and returns true, or false on eof (and closes the file).
// It ignores the utf8 file signature ByteOrderMark (U+FEFF=EF BB BF), checks
// for valid utf-8 and allows space or tab between fields.
// utf8_str is set with the unichar string, and bounding box with the box.
// If there are page numbers in the file, it reads them all.
bool ReadNextBox(int *line_number, FILE* box_file,
STRING* utf8_str, TBOX* bounding_box) {
return ReadNextBox(-1, line_number, box_file, utf8_str, bounding_box);
}
