Boxes define boundaries for "blobs" or "word lines". A box file contains box definitions for blobs or words, one per line of the file. The can be used to train tesseract to recognize new character sets, particularly when training for new languages.
From applybox.cpp in the source code:
The box file is assumed to contain box definitions, one per line, of the following format for blob-level boxes: and for word/line-level boxes: WordStr # NOTES: The boxes use tesseract coordinates, i.e. 0,0 is at BOTTOM-LEFT.
is 0-based, and the page number is used for multipage input (tiff).
In the blob-level form, each line represents a recognizable unit, which may be several UTF-8 bytes, but there is a bounding box around each recognizable unit, and no classifier is needed to train in this mode (bootstrapping.)
In the word/line-level form, the line begins with the literal "WordStr", and the bounding box bounds either a whole line or a whole word. The recognizable units in the word/line are listed after the # at the end of the line and are space delimited, ignoring any original spaces on the line. Eg. word -> #w o r d multi word line -> #m u l t i w o r d l i n e The recognizable units must be space-delimited in order to allow multiple unicodes to be used for a single recognizable unit, eg Hindi. In this mode, the classifier must have been pre-trained with the desired character set, or it will not be able to find the character segmentations.
Tesseract can create a boxfile for a given image.
In versions of tesseract before 3.0.3 you must give the name of the image file, the base name for the output file and word makebox, which is the name of a configuration file that contains the parameter tessedit_create_boxfile 1. In TrainingTesseract3 an additional configuration file argument, batch.nochop is used, and that file contains the parameters chop_enable 0 and wordrec_enable_assoc 0.
So a full command would look something like:
tesseract name-of-pdf-img.ppm name-of-outputfile batch.nochop makeboxjTessBoxEditor is a Java GUI application which takes a boxfile and a tiff image and provides a visual editor for editing the boxfile. This allows you to make corrections to the characters tesseract identified.
Part of TrainingTesseract3, but can you do just the boxfile training and create an alternative tessdata for a language?
tesseract [lang].[fontname].exp[num].tif [lang].[fontname].exp[num] box.trainThis should generate a [lang].[fontname].exp[num].tr and a [lang].[fontname].exp[num].txt file.
The box.train config sets the following parameters:
disable_character_fragments T
file_type .bl
textord_fast_pitch_test T
tessedit_single_match 0
tessedit_zero_rejection T
tessedit_minimal_rejection F
tessedit_write_rep_codes F
il1_adaption_test 1
edges_children_fix F
edges_childarea 0.65
edges_boxarea 0.9
tessedit_resegment_from_boxes T
tessedit_train_from_boxes T
textord_no_rejects T