ocr/ocr_pyflow

#!/usr/bin/env python2.7
# coding=utf-8



"""
ocr_pyflow.py

Date:   18/10/2018
Usage:  For usage instructions run with option --help
Author: Madis Rumming <mrumming@uni-bielefeld.de>
"""

__author__ = "Madis Rumming <mrumming@uni-bielefeld.de>"
__copyright__ = "Copyright 2018, Data Infrastructure and Digital Humanities,\
SFB 1288, Bielefeld University"

__version__ = "0.7"
__maintainer__ = "Patrick Jentsch"
__email__ = "p.jentsch@uni-bielefeld.de"
__status__ = "Development"



import argparse
import multiprocessing
import os
import sys
import unicodedata
from pyflow import WorkflowRunner
from shlex import shlex



''' TODO:
  ' Implement --end-page: Last page to ocr
  ' Implement --memMb: Total amount of memory (RAM) available for this workflow. Default: 2048 * nCores
  ' 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("Performs OCR of (historical) documents utilizing OCRopus for preprocessing and Tesseract OCR \
        for OCR. Available outputs are HOCR, PDF, shrinked PDF, and simple DTAbf \
        (TEI P5 compliant). Software requirements: imagemagick, ocropus, pdftk, pdftoppm, poppler-utils, pyflow, python2.7, tesseract")

    parser.add_argument("-i",
                        dest="inputDir",
                        help="Input directory.",
                        required=True)
    parser.add_argument("-l",
                        choices=["deu", "deu_frak", "eng", "enm", "fra", "spa", "frm"],
                        dest='lang',
                        help="Language for OCR",
                        required=True)
    parser.add_argument("-o",
                        dest="outputDir",
                        help="Output directory.",
                        required=True)
    parser.add_argument("--skip-image",
                        action='store_true',
                        default=False,
                        dest="skip_images",
                        help="Skip detection of images as input.",
                        required=False)
    parser.add_argument("--skip-pdf",
                        action='store_true',
                        default=False,
                        dest="skip_pdf",
                        help="Skip detection of PDFs as input.",
                        required=False)
    parser.add_argument("--pdf",
                        action='store_true',
                        default=False,
                        dest='pdf',
                        help="Create PDF files.",
                        required=False)
    parser.add_argument("--keep-intermediates",
                        action='store_true',
                        default=False,
                        dest="keepIntermediates",
                        help="Keep intermediate files.",
                        required=False)
    parser.add_argument("--nCores",
                        default=multiprocessing.cpu_count(),
                        dest="nCores",
                        help="Total number of cores available.",
                        required=False,
                        type=int)
    return parser.parse_args()



class OCRWorkflow(WorkflowRunner):
    def __init__(self, jobs, lang, pdf, keepIntermediates, nCores):
        self.jobs = jobs
        self.lang = lang
        self.pdf = pdf
        self.keepIntermediates = keepIntermediates
        self.nCores = nCores



    def workflow(self):
        ###
        # Task "mkdir_job": create output directories
        # Dependencies: None
        ###
        mkdir_jobs = []
        mkdir_job_number = 0
        for job in self.jobs["images"] + self.jobs["pdf"]:
            mkdir_job_number += 1
            cmd = "mkdir -p %s %s %s %s" % (
                os.path.join(job["output_dir"], "hocr_files"),
                os.path.join(job["output_dir"], "tmp", "ocropus-nlbin"),
                os.path.join(job["output_dir"], "tmp", "tesseract"),
                os.path.join(job["output_dir"], "tmp", "tiff_files"))
            cmd = shlex().escape(cmd);
            mkdir_jobs.append(self.addTask(label="mkdir_job_-_%i" % (mkdir_job_number), command=cmd))


        ###
        # Task "split_job": split input file into one tiff file per page
        # Dependencies: mkdir_jobs
        ###
        split_jobs = []
        split_job_number = 0
        for job in self.jobs["images"]:
            split_job_number += 1
            # TODO: Make the following command work
            '''
            cmd = "convert %s %s" % (
                job["path"],
                os.path.join(job["output_dir"], "tmp", "tiff_files", os.path.basename(job["path"]).rsplit(".", 1)[0] + "-%sd.tif" % ("%")))
            '''
            # WORKAROUND
            cmd = "tiff2pdf -o %s %s && pdftoppm %s %s -tiff -r 300 -tiffcompression lzw -cropbox && rm %s" % (
                os.path.join(job["output_dir"], "tmp", "tiff_files", os.path.basename(job["path"]).rsplit(".", 1)[0] + ".pdf"),
                job["path"],
                os.path.join(job["output_dir"], "tmp", "tiff_files", os.path.basename(job["path"]).rsplit(".", 1)[0] + ".pdf"),
                os.path.join(job["output_dir"], "tmp", "tiff_files", os.path.basename(job["path"]).rsplit(".", 1)[0]),
                os.path.join(job["output_dir"], "tmp", "tiff_files", os.path.basename(job["path"]).rsplit(".", 1)[0] + ".pdf"))
            cmd = shlex().escape(cmd);
            split_jobs.append(self.addTask(label="split_job_-_%i" % (split_job_number), command=cmd, dependencies=mkdir_jobs))
        for job in self.jobs["pdf"]:
            split_job_number += 1
            cmd = "pdftoppm %s %s -tiff -r 300 -tiffcompression lzw -cropbox" % (
                job["path"],
                os.path.join(job["output_dir"], "tmp", "tiff_files", os.path.basename(job["path"]).rsplit(".", 1)[0]))
            cmd = shlex().escape(cmd);
            split_jobs.append(self.addTask(label="split_job_-_%i" % (split_job_number), command=cmd, dependencies=mkdir_jobs))


        ###
        # Task "ocropus_nlbin_job": binarize tiff files from previous split
        # Dependencies: split_jobs
        ###
        ocropusnlbin_jobs = []
        ocropusnlbin_job_number = 0
        for job in self.jobs["images"] + self.jobs["pdf"]:
            ocropusnlbin_job_number += 1
            cmd = "ocropus-nlbin -o %s %s" % (
                os.path.join(job["output_dir"], "tmp", "ocropus-nlbin"),
                os.path.join(job["output_dir"], "tmp", "tiff_files", os.path.basename(job["path"]).rsplit(".", 1)[0] + "-*.tif"))
            cmd = shlex().escape(cmd);
            ocropusnlbin_jobs.append(self.addTask(label="ocropusnlbin_job_-_%i" % (ocropusnlbin_job_number), command=cmd, dependencies=split_jobs))


        ###
        # Task "tesseract_job": perform OCR on binarized images
        # Dependencies: ocropusnlbin_jobs
        ###
        self.waitForTasks()
        tesseract_jobs = []
        tesseract_job_number = 0
        for job in self.jobs["images"] + self.jobs["pdf"]:
            # This list is empty if you don't wait for ocropus_nlbin_jobs to complete
            for file in filter(lambda x: x.endswith(".bin.png"), os.listdir(os.path.join(job["output_dir"], "tmp", "ocropus-nlbin"))):
                tesseract_job_number += 1
                cmd = "tesseract %s %s -l %s hocr %s" % (
                    os.path.join(job["output_dir"], "tmp", "ocropus-nlbin", file),
                    os.path.join(job["output_dir"], "tmp", "tesseract", file.rsplit(".", 2)[0]),
                    self.lang,
                    "pdf" if self.pdf else "")
                cmd = shlex().escape(cmd);
                tesseract_jobs.append(self.addTask(label="tesseract_job_-_%i" % (tesseract_job_number), command=cmd, dependencies=ocropusnlbin_jobs, nCores=min(4, self.nCores)))


        ###
        # Task "pdf_merge_job": Merge PDF files
        # Dependencies: tesseract_jobs
        ###
        pdf_merge_jobs = []
        pdf_merge_job_number = 0
        if self.pdf:
            for job in self.jobs["images"] + self.jobs["pdf"]:
                pdf_merge_job_number += 1
                cmd = "pdftk %s cat output %s" % (
                    os.path.join(job["output_dir"], "tmp", "tesseract", "*.pdf"),
                    os.path.join(job["output_dir"], os.path.basename(job["path"].rsplit(".", 1)[0] + ".pdf")))
                cmd = shlex().escape(cmd);
                pdf_merge_jobs.append(self.addTask(label="pdf_merge_job_-_%i" % (pdf_merge_job_number), command=cmd, dependencies=tesseract_jobs))


        ###
        # Task "pdf_to_txt_jobs":
        # Dependencies: pdf_merge_jobs
        ###
        pdf_to_txt_jobs = []
        pdf_to_txt_job_number = 0
        if self.pdf:
            for job in self.jobs["images"] + self.jobs["pdf"]:
                pdf_to_txt_job_number += 1
                cmd = "pdftotext -raw %s" % (
                    os.path.join(job["output_dir"], os.path.basename(job["path"].rsplit(".", 1)[0] + ".pdf")))
                cmd = shlex().escape(cmd);
                pdf_merge_jobs.append(self.addTask(label="pdf_to_txt_job_-_%i" % (pdf_to_txt_job_number), command=cmd, dependencies=pdf_merge_jobs))


        ###
        # Task "move_hocr_job": move hocr files from <output_dir>/tmp/tesseract to <output_dir>/hocr_files
        # Dependencies: tesseract_jobs
        ###
        move_hocr_jobs = []
        move_hocr_job_number = 0
        for job in self.jobs["images"] + self.jobs["pdf"]:
            move_hocr_job_number += 1
            cmd = "mv %s %s" % (
                os.path.join(job["output_dir"], "tmp", "tesseract", "*.hocr"),
                os.path.join(job["output_dir"], "hocr_files"))
            cmd = shlex().escape(cmd);
            move_hocr_jobs.append(self.addTask(label="move_hocr_job_-_%i" % (move_hocr_job_number), command=cmd, dependencies=tesseract_jobs))


        ###Total amount of memory (RAM) available for this workflow. Default: 2048 * nCores"
        # Task "hocr_to_teip5_job": create TEI P5 file from hocr files
        # Dependencies: move_hocr_jobs
        ###
        hocr_to_teip5_jobs = []
        hocr_to_teip5_job_number = 0
        for job in self.jobs["images"] + self.jobs["pdf"]:
            hocr_to_teip5_job_number += 1
            cmd = "parse_hocr %s %s" % (
                os.path.join(job["output_dir"], "hocr_files"),
                os.path.join(os.path.join(job["output_dir"], os.path.basename(job["path"]).rsplit(".", 1)[0] + ".xml")))
            cmd = shlex().escape(cmd);
            hocr_to_teip5_jobs.append(self.addTask(label="hocr_to_teip5_job_-_%i" % (hocr_to_teip5_job_number), command=cmd, dependencies=move_hocr_jobs))


        ###
        # Task "cleanup_job": remove temporary files
        # Dependencies: All
        ###
        self.waitForTasks()
        cleanup_jobs = []
        cleanup_job_counter = 0
        if not self.keepIntermediates:
            for job in self.jobs["images"] + self.jobs["pdf"]:
                cleanup_job_counter += 1
                cmd = "rm -r %s" % (os.path.join(job["output_dir"], "tmp"))
                cmd = shlex().escape(cmd);
                cleanup_jobs.append(self.addTask(label="cleanup_job_-_%i" % (cleanup_job_counter), command=cmd))



def analyze_jobs(inputDir, outputDir, skip_pdf=False, skip_images=False):
    files = os.listdir(inputDir)
    images = []
    pdf = []

    if not skip_images:
        for file in filter(lambda x: x.endswith(".tif") or x.endswith(".tiff"), files):
            images.append({"path": os.path.join(inputDir, file), "output_dir": os.path.join(outputDir, file.rsplit(".", 1)[0])})
    if not skip_pdf:
        for file in filter(lambda x: x.endswith(".pdf"), files):
            pdf.append({"path": os.path.join(inputDir, file), "output_dir": os.path.join(outputDir, file.rsplit(".", 1)[0])})
    for subDir in filter(lambda x: os.path.isdir(os.path.join(inputDir, x)), files):
        subDirFiles = os.listdir(os.path.join(inputDir, subDir))
        if not skip_images:
            for file in filter(lambda x: x.endswith(".tif") or x.endswith(".tiff"), subDirFiles):
                images.append({"path": os.path.join(inputDir, subDir, file), "output_dir": os.path.join(outputDir, subDir, file.rsplit(".", 1)[0])})
        if not skip_pdf:
            for file in filter(lambda x: x.endswith(".pdf"), subDirFiles):
                pdf.append({"path": os.path.join(inputDir, subDir, file), "output_dir": os.path.join(outputDir, subDir, file.rsplit(".", 1)[0])})
    return {"pdf": pdf, "images": images}



def normalize_input_filenames(path):
    ###
    # Normalize input filenames and directories to avoid bugs and also for better usage and readability.
    ###
    for file in os.listdir(path):
        file_with_path = os.path.join(path, file)
        if os.path.isdir(file_with_path):
            normalize_input_filenames(file_with_path)
        new_file_with_path = os.path.join(path, unicodedata.normalize("NFKD", file.decode("utf-8")).encode("ascii", "ignore").replace(" ", "_"))
        os.rename(file_with_path, new_file_with_path)



def main():
    args = parse_arguments()

    normalize_input_filenames(args.inputDir)
    jobs = analyze_jobs(args.inputDir, args.outputDir, skip_pdf=args.skip_pdf, skip_images=args.skip_images)

    wflow = OCRWorkflow(jobs, args.lang, args.pdf, args.keepIntermediates, args.nCores)
    retval = wflow.run(nCores=args.nCores)
    sys.exit(retval)



if __name__ == "__main__":
    main()