// Reads the next box from the given box file into TBOX.
bool read_b(int applybox_page, int *line_number, FILE *box_file,
char *label, TBOX *bbox) {
int x_min, y_min, x_max, y_max;
if (read_next_box(applybox_page, line_number, box_file, label,
&x_min, &y_min, &x_max, &y_max)) {
bbox->set_to_given_coords(x_min, y_min, x_max, y_max);
return true;
} else {
return false;
}
}
void apply_boxes(BLOCK_LIST *block_list //real blocks
) {
INT16 boxfile_lineno = 0;
INT16 boxfile_charno = 0;
BOX box; //boxfile box
char ch[2]; //correct ch from boxfile
ROW *row;
ROW *prev_row = NULL;
INT16 prev_box_right = MAX_INT16;
INT16 block_id;
INT16 row_id;
INT16 box_count = 0;
INT16 box_failures = 0;
INT16 labels_ok;
INT16 rows_ok;
INT16 bad_blobs;
INT16 tgt_char_counts[128]; //No. of box samples
// INT16 labelled_char_counts[128]; //No. of unique labelled samples
INT16 i;
INT16 rebalance_count = 0;
char min_char;
INT16 min_samples;
INT16 final_labelled_blob_count;
for (i = 0; i < 128; i++)
tgt_char_counts[i] = 0;
FILE* box_file;
STRING filename = imagefile;
filename += ".box";
if (!(box_file = fopen (filename.string(), "r"))) {
CANTOPENFILE.error ("read_next_box", EXIT,
"Cant open box file %s %d",
filename.string(), errno);
}
ch[1] = '\0';
clear_any_old_text(block_list);
while (read_next_box (box_file, &box, &ch[0])) {
box_count++;
tgt_char_counts[ch[0]]++;
row = find_row_of_box (block_list, box, block_id, row_id);
if (box.left () < prev_box_right) {
boxfile_lineno++;
boxfile_charno = 1;
}
else
boxfile_charno++;
if (row == NULL) {
box_failures++;
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"FAILURE! box overlaps no blobs or blobs in multiple rows");
}
else {
if ((box.left () >= prev_box_right) && (row != prev_row))
report_failed_box (boxfile_lineno, boxfile_charno, box, ch,
"WARNING! false row break");
box_failures += resegment_box (row, box, ch, block_id, row_id,
boxfile_lineno, boxfile_charno);
prev_row = row;
}
prev_box_right = box.right ();
}
tidy_up(block_list,
labels_ok,
rows_ok,
bad_blobs,
tgt_char_counts,
rebalance_count,
min_char,
min_samples,
final_labelled_blob_count);
tprintf ("APPLY_BOXES:\n");
tprintf (" Boxes read from boxfile: %6d\n", box_count);
tprintf (" Initially labelled blobs: %6d in %d rows\n",
labels_ok, rows_ok);
tprintf (" Box failures detected: %6d\n", box_failures);
tprintf (" Duped blobs for rebalance:%6d\n", rebalance_count);
tprintf (" \"%c\" has fewest samples:%6d\n", min_char, min_samples);
tprintf (" Total unlabelled words: %6d\n",
bad_blobs);
tprintf (" Final labelled words: %6d\n",
final_labelled_blob_count);
}
// Box files are used ONLY DURING TRAINING, but by both processes of
// creating tr files with tesseract, and unicharset_extractor.
// read_next_box 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 uft8 file signature, checks for valid utf-8 and allows
// space or tab between fields.
// utf8_str must be at least kBoxReadBufSize in length.
// If there are page numbers in the file, it reads them all.
bool read_next_box(int *line_number, FILE* box_file, char* utf8_str,
int* x_min, int* y_min, int* x_max, int* y_max) {
return read_next_box(-1, line_number, box_file, utf8_str,
x_min, y_min, x_max, y_max);
}
bool read_next_box(FILE* box_file, char* utf8_str,
int* x_min, int* y_min, int* x_max, int* y_max) {
return read_next_box(-1, box_file, utf8_str,
x_min, y_min, x_max, y_max);
}
