mirror of
https://gitlab.ub.uni-bielefeld.de/sfb1288inf/ocr.git
synced 2025-01-13 22:20:34 +00:00
500 lines
18 KiB
Python
Executable File
500 lines
18 KiB
Python
Executable File
#!/usr/bin/env python2.7
|
|
# coding=utf-8
|
|
|
|
|
|
"""
|
|
ocr
|
|
|
|
Usage: For usage instructions run with option --help
|
|
Author: Patrick Jentsch <p.jentsch@uni-bielefeld.de>
|
|
"""
|
|
|
|
|
|
import argparse
|
|
import multiprocessing
|
|
import os
|
|
import re
|
|
import sys
|
|
from pyflow import WorkflowRunner
|
|
|
|
|
|
''' TODO:
|
|
' Implement --end-page: Last page to ocr
|
|
' Implement --memMb: Total amount of memory (RAM) available for this workflow.
|
|
' Default: 2048 * n_cores
|
|
' Implement --rotate: Rotate pages from input (90, 180, 270)
|
|
' Implement --split-pages: Split pages in half after possible rotation
|
|
' Implement --start-page: First page to ocr
|
|
'''
|
|
|
|
|
|
def parse_arguments():
|
|
parser = argparse.ArgumentParser(
|
|
description='Performs OCR of (historical) documents utilizing OCRopus for preprocessing and Tesseract OCR for OCR. The results are served as hOCR, PDF, raw text and TEI compliant XML files.\nSoftware requirements: imagemagick, ocropus, pdftoppm, pdfunite, poppler-utils, pyflow, python2.7, python3.5, tesseract'
|
|
)
|
|
parser.add_argument(
|
|
'-i',
|
|
dest='input_dir',
|
|
required=True
|
|
)
|
|
parser.add_argument(
|
|
'-l',
|
|
choices=[
|
|
'deu', 'deu_frak', 'eng', 'enm', 'fra', 'frm', 'ita', 'por', 'spa'
|
|
],
|
|
dest='lang',
|
|
required=True
|
|
)
|
|
parser.add_argument(
|
|
'-o',
|
|
dest='output_dir',
|
|
required=True
|
|
)
|
|
parser.add_argument(
|
|
'--skip-binarisation',
|
|
action='store_true',
|
|
default=False,
|
|
dest='skip_binarisation',
|
|
help='skip ocropy binarisation',
|
|
required=False
|
|
)
|
|
parser.add_argument(
|
|
'--keep-intermediates',
|
|
action='store_true',
|
|
default=False,
|
|
dest='keep_intermediates',
|
|
help='keep intermediate files',
|
|
required=False
|
|
)
|
|
parser.add_argument(
|
|
'--nCores',
|
|
default=min(4, multiprocessing.cpu_count()),
|
|
dest='n_cores',
|
|
help='total number of cores available',
|
|
required=False,
|
|
type=int
|
|
)
|
|
return parser.parse_args()
|
|
|
|
|
|
class OCRWorkflow(WorkflowRunner):
|
|
def __init__(self, args):
|
|
self.jobs = analyze_jobs(args.input_dir, args.output_dir)
|
|
self.skip_binarisation = args.skip_binarisation
|
|
self.keep_intermediates = args.keep_intermediates
|
|
self.lang = args.lang
|
|
self.n_cores = args.n_cores
|
|
|
|
def workflow(self):
|
|
if len(self.jobs) == 0:
|
|
return
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Create output directories #
|
|
' ##################################################
|
|
'''
|
|
create_output_directories_jobs = []
|
|
for index, job in enumerate(self.jobs):
|
|
cmd = 'mkdir -p "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp')
|
|
)
|
|
if self.keep_intermediates:
|
|
cmd += ' "%s" "%s" "%s" "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp', 'hocr'),
|
|
os.path.join(job['output_dir'], 'tmp', 'pdf'),
|
|
os.path.join(job['output_dir'], 'tmp', 'tiff'),
|
|
os.path.join(job['output_dir'], 'tmp', 'txt')
|
|
)
|
|
if not self.skip_binarisation:
|
|
cmd += ' "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp', 'bin.png')
|
|
)
|
|
create_output_directories_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
label='create_output_directories_job_-_%i' % (index)
|
|
)
|
|
)
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Split #
|
|
' ##################################################
|
|
'''
|
|
split_jobs = []
|
|
split_job_n_cores = min(
|
|
self.n_cores,
|
|
max(1, int(self.n_cores / len(self.jobs)))
|
|
)
|
|
for index, job in enumerate(self.jobs):
|
|
if job['filename'].endswith(('.tif', '.tiff')):
|
|
'''
|
|
' This command also works for PDF input but ocropus-nlbin
|
|
' is not able to handle the TIFF output of it.
|
|
'''
|
|
cmd = 'convert -density 300 "%s" -compress LZW -scene 1 "%s/page-%%d.tif"' % (
|
|
job['path'],
|
|
os.path.join(job['output_dir'], 'tmp')
|
|
)
|
|
else:
|
|
cmd = 'pdftoppm -r 300 -tiff -tiffcompression lzw "%s" "%s"' % (
|
|
job['path'],
|
|
os.path.join(job['output_dir'], 'tmp', 'page')
|
|
)
|
|
|
|
split_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies='create_output_directories_job_-_%i' % (index),
|
|
label='split_job_-_%i' % (index),
|
|
nCores=split_job_n_cores
|
|
)
|
|
)
|
|
|
|
if not self.skip_binarisation:
|
|
'''
|
|
' The binarisation_jobs are created based of the output files of
|
|
' the split_jobs. So wait until they are finished.
|
|
'''
|
|
self.waitForTasks()
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Binarise #
|
|
' ##################################################
|
|
'''
|
|
binarisation_jobs = []
|
|
'''
|
|
' We run ocropus-nlbin with either four or, if there are less then
|
|
' four cores available for this workflow, the available core
|
|
' number.
|
|
'''
|
|
binarisation_job_n_cores = min(4, self.n_cores)
|
|
for index, job in enumerate(self.jobs):
|
|
files = os.listdir(os.path.join(job['output_dir'], 'tmp'))
|
|
files = filter(lambda x: x.endswith('.tif'), files)
|
|
files.sort(key=lambda x: int(re.search(r'\d+', x).group(0)))
|
|
files = map(
|
|
lambda x: '"' + os.path.join(job['output_dir'], 'tmp', x) + '"',
|
|
files
|
|
)
|
|
cmd = 'ocropus-nlbin --output "%s" --parallel "%i" %s' % (
|
|
os.path.join(job['output_dir'], 'tmp'),
|
|
binarisation_job_n_cores,
|
|
' '.join(files)
|
|
)
|
|
binarisation_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies='split_job_-_%i' % (index),
|
|
label='binarisation_job_-_%i' % (index),
|
|
nCores=binarisation_job_n_cores
|
|
)
|
|
)
|
|
|
|
'''
|
|
' The post_binarisation_jobs are created based of the output files
|
|
' of the binarisation_jobs. So wait until they are finished.
|
|
'''
|
|
self.waitForTasks()
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Normalise file names from binarisation #
|
|
' ##################################################
|
|
'''
|
|
post_binarisation_jobs = []
|
|
for index, job in enumerate(self.jobs):
|
|
number = 0
|
|
files = os.listdir(os.path.join(job['output_dir'], 'tmp'))
|
|
files = filter(lambda x: x.endswith('.bin.png'), files)
|
|
files.sort()
|
|
for file in files:
|
|
cmd = 'mv "%s" "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp', file),
|
|
os.path.join(
|
|
job['output_dir'],
|
|
'tmp',
|
|
'page-%i.bin.png' % (int(file.split('.', 1)[0]))
|
|
)
|
|
)
|
|
post_binarisation_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies='binarisation_job_-_%i' % (index),
|
|
label='post_binarisation_job_-_%i-%i' % (
|
|
index,
|
|
number
|
|
)
|
|
)
|
|
)
|
|
number += 1
|
|
|
|
'''
|
|
' The ocr_jobs are created based of the output files of either the
|
|
' split_jobs or post_binarisation_jobs. So wait until they are
|
|
' finished.
|
|
'''
|
|
self.waitForTasks()
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Optical Character Recognition #
|
|
' ##################################################
|
|
'''
|
|
ocr_jobs = []
|
|
'''
|
|
' Tesseract runs fastest with four cores. So we run it with either four
|
|
' or, if there are less then four cores available for this workflow,
|
|
' the available core number.
|
|
'''
|
|
ocr_job_n_cores = min(4, self.n_cores)
|
|
'''
|
|
' WORKAROUND: Tesseract only uses one core for the deu_frak language
|
|
' model, so the workflow will also only reserve one in this case.
|
|
'''
|
|
if self.lang == "deu_frak":
|
|
ocr_job_n_cores = 1
|
|
for index, job in enumerate(self.jobs):
|
|
files = os.listdir(os.path.join(job['output_dir'], 'tmp'))
|
|
if self.skip_binarisation:
|
|
files = filter(lambda x: x.endswith('.tif'), files)
|
|
else:
|
|
files = filter(lambda x: x.endswith('.bin.png'), files)
|
|
files.sort(key=lambda x: int(re.search(r'\d+', x).group(0)))
|
|
files = map(
|
|
lambda x: os.path.join(job['output_dir'], 'tmp', x),
|
|
files
|
|
)
|
|
number = 0
|
|
for file in files:
|
|
cmd = 'tesseract "%s" "%s" -l "%s" hocr pdf txt' % (
|
|
file,
|
|
os.path.join(
|
|
job['output_dir'],
|
|
'tmp',
|
|
file.rsplit('.', 1 if self.skip_binarisation else 2)[0]
|
|
),
|
|
self.lang
|
|
)
|
|
if self.skip_binarisation:
|
|
ocr_job_dependencies = 'split_job_-_%i' % (index)
|
|
else:
|
|
ocr_job_dependencies = filter(
|
|
lambda x: x == 'post_binarisation_job_-_%i-%i' % (
|
|
index,
|
|
number
|
|
),
|
|
post_binarisation_jobs
|
|
)
|
|
ocr_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies=ocr_job_dependencies,
|
|
label='ocr_job_-_%i-%i' % (index, number),
|
|
nCores=ocr_job_n_cores
|
|
)
|
|
)
|
|
number += 1
|
|
|
|
'''
|
|
' The following jobs are created based of the output files of the
|
|
' ocr_jobs. So wait until they are finished.
|
|
'''
|
|
self.waitForTasks()
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Create TEI P5 files #
|
|
' ##################################################
|
|
'''
|
|
hocr_to_tei_jobs = []
|
|
for index, job in enumerate(self.jobs):
|
|
files = os.listdir(os.path.join(job['output_dir'], 'tmp'))
|
|
files = filter(lambda x: x.endswith('.hocr'), files)
|
|
files.sort(key=lambda x: int(re.search(r'\d+', x).group(0)))
|
|
files = map(
|
|
lambda x: '"' + os.path.join(job['output_dir'], 'tmp', x) + '"',
|
|
files
|
|
)
|
|
cmd = 'hocrtotei %s "%s"' % (
|
|
' '.join(files),
|
|
os.path.join(
|
|
job['output_dir'],
|
|
os.path.join(job['output_dir'], job['name'] + '.xml')
|
|
)
|
|
)
|
|
hocr_to_tei_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies=filter(
|
|
lambda x: x.startswith('ocr_job_-_%i' % (index)),
|
|
ocr_jobs
|
|
),
|
|
label='hocr_to_tei_job_-_%i' % (index)
|
|
)
|
|
)
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Merge PDF files #
|
|
' ##################################################
|
|
'''
|
|
pdf_merge_jobs = []
|
|
for index, job in enumerate(self.jobs):
|
|
files = os.listdir(os.path.join(job['output_dir'], 'tmp'))
|
|
files = filter(lambda x: x.endswith('.pdf'), files)
|
|
files.sort(key=lambda x: int(re.search(r'\d+', x).group(0)))
|
|
files = map(
|
|
lambda x: '"' + os.path.join(job['output_dir'], 'tmp', x) + '"',
|
|
files
|
|
)
|
|
cmd = 'pdfunite %s "%s"' % (
|
|
' '.join(files),
|
|
os.path.join(
|
|
job['output_dir'],
|
|
os.path.join(job['output_dir'], job['name'] + '.pdf')
|
|
)
|
|
)
|
|
pdf_merge_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies=filter(
|
|
lambda x: x.startswith('ocr_job_-_%i' % (index)),
|
|
ocr_jobs
|
|
),
|
|
label='pdf_merge_job_-_%i' % (index)
|
|
)
|
|
)
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Merge text files #
|
|
' ##################################################
|
|
'''
|
|
txt_merge_jobs = []
|
|
for index, job in enumerate(self.jobs):
|
|
files = os.listdir(os.path.join(job['output_dir'], 'tmp'))
|
|
files = filter(lambda x: x.endswith('.txt'), files)
|
|
files.sort(key=lambda x: int(re.search(r'\d+', x).group(0)))
|
|
files = map(
|
|
lambda x: '"' + os.path.join(job['output_dir'], 'tmp', x) + '"',
|
|
files
|
|
)
|
|
cmd = 'cat %s > "%s"' % (
|
|
' '.join(files),
|
|
os.path.join(
|
|
job['output_dir'],
|
|
os.path.join(job['output_dir'], job['name'] + '.txt')
|
|
)
|
|
)
|
|
txt_merge_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies=filter(
|
|
lambda x: x.startswith('ocr_job_-_%i' % (index)),
|
|
ocr_jobs
|
|
),
|
|
label='txt_merge_job_-_%i' % (index)
|
|
)
|
|
)
|
|
|
|
'''
|
|
' ##################################################
|
|
' # Cleanup #
|
|
' ##################################################
|
|
'''
|
|
cleanup_jobs = []
|
|
if self.keep_intermediates:
|
|
for index, job in enumerate(self.jobs):
|
|
cleanup_job_dependencies = [
|
|
'hocr_to_tei_job_-_%i' % (index),
|
|
'pdf_merge_job_-_%i' % (index),
|
|
'txt_merge_job_-_%i' % (index)
|
|
]
|
|
cmd = 'mv "%s"/*.hocr "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp'),
|
|
os.path.join(job['output_dir'], 'tmp', 'hocr'),
|
|
)
|
|
cmd += ' && mv "%s"/*.pdf "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp'),
|
|
os.path.join(job['output_dir'], 'tmp', 'pdf'),
|
|
)
|
|
cmd += ' && mv "%s"/*.tif "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp'),
|
|
os.path.join(job['output_dir'], 'tmp', 'tiff'),
|
|
)
|
|
cmd += ' && mv "%s"/*.txt "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp'),
|
|
os.path.join(job['output_dir'], 'tmp', 'txt'),
|
|
)
|
|
if not self.skip_binarisation:
|
|
cmd += ' && mv "%s"/*.bin.png "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp'),
|
|
os.path.join(job['output_dir'], 'tmp', 'bin.png'),
|
|
)
|
|
cmd += ' && rm "%s"/*.nrm.png' % (
|
|
os.path.join(job['output_dir'], 'tmp')
|
|
)
|
|
cleanup_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies=cleanup_job_dependencies,
|
|
label='cleanup_job_-_%i' % (index)
|
|
)
|
|
)
|
|
else:
|
|
for index, job in enumerate(self.jobs):
|
|
cleanup_job_dependencies = [
|
|
'hocr_to_tei_job_-_%i' % (index),
|
|
'pdf_merge_job_-_%i' % (index),
|
|
'txt_merge_job_-_%i' % (index)
|
|
]
|
|
cmd = 'rm -r "%s"' % (
|
|
os.path.join(job['output_dir'], 'tmp')
|
|
)
|
|
cleanup_jobs.append(
|
|
self.addTask(
|
|
command=cmd,
|
|
dependencies=cleanup_job_dependencies,
|
|
label='cleanup_job_-_%i' % (index)
|
|
)
|
|
)
|
|
|
|
|
|
def analyze_jobs(input_dir, output_dir):
|
|
jobs = []
|
|
|
|
for file in os.listdir(input_dir):
|
|
if os.path.isdir(os.path.join(input_dir, file)):
|
|
jobs += analyze_jobs(
|
|
os.path.join(input_dir, file),
|
|
os.path.join(output_dir, file)
|
|
)
|
|
elif file.endswith(('.pdf', '.tif', '.tiff')):
|
|
jobs.append(
|
|
{
|
|
'filename': file,
|
|
'name': file.rsplit('.', 1)[0],
|
|
'output_dir': os.path.join(output_dir, file),
|
|
'path': os.path.join(input_dir, file)
|
|
}
|
|
)
|
|
|
|
return jobs
|
|
|
|
|
|
def main():
|
|
args = parse_arguments()
|
|
|
|
wflow = OCRWorkflow(args)
|
|
|
|
retval = wflow.run(dataDirRoot=args.output_dir, nCores=args.n_cores)
|
|
|
|
sys.exit(retval)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|