ocrd-galley/my_ocrd_workflow

#!/bin/bash
LOG_LEVEL='DEBUG'


set -e  # Abort on error
if [ "$LOG_LEVEL" = "DEBUG" -o "$LOG_LEVEL" = "TRACE" ]; then
  set -x
fi


remove_filegrp() {
  # Remove the given file group from the workspace

  filegrp_use=$1
  mets=$2

  xmlstarlet ed --inplace \
    -N mets=http://www.loc.gov/METS/ \
    -d "//mets:fileGrp[@USE='$filegrp_use']" $mets

  # XXX See also https://github.com/OCR-D/core/issues/245
  # XXX This should also delete the files (after checking if they are indeed inside the workspace) and the directory
}

do_binarization() {
  # Binarize the images

  remove_filegrp OCR-D-IMG-BIN mets.xml
  ocrd-kraken-binarize -l $LOG_LEVEL \
    -m mets.xml -I OCR-D-IMG -O OCR-D-IMG-BIN
}

do_fontident() {
  # Identify fonts in the images

  network=`python3 -c "import ocrd_typegroups_classifier, os; print(os.path.join(os.path.dirname(ocrd_typegroups_classifier.__file__), 'models', 'classifier.tgc'))"`
  ocrd_typegroups_classifier_parameters="
  {
    \"network\": \"$network\",
    \"stride\":  143
  }"

  remove_filegrp OCR-D-FONTIDENT mets.xml
  ocrd-typegroups-classifier -l $LOG_LEVEL \
    -m mets.xml -I OCR-D-IMG -O OCR-D-FONTIDENT \
    -p <(echo $ocrd_typegroups_classifier_parameters)
  # XXX Check if ocrd-typegroups-classifier uses the whole image
  # XXX does DEFAULT have any meaning? /buerger_gedichte_1778.ocrd does not have
  #     any DEFAULT, yet -I DEFAULT seems to work for ocrd-typegroups-classifier
  #ocrd workspace validate mets.xml
  # XXX <error>Unspecified USE category 'FONTIDENT' in fileGrp 'OCR-D-FONTIDENT'</error>
  # XXX <error>File 'OCR-D-FONTIDENT_OCR-D-IMG_0002' does not manifest any physical page.</error>
  # XXX <notice>Won't download remote image <OCR-D-IMG/OCR-D-IMG_0002></notice>
}

do_linesegmentation() {
  # Segment the lines in the binarized images

  remove_filegrp OCR-D-SEG-REGION mets.xml
  remove_filegrp OCR-D-SEG-LINE mets.xml
  #ocrd-ocropy-segment -l $LOG_LEVEL \
  #  -m mets.xml -I OCR-D-IMG-BIN -O OCR-D-SEG-LINE
  # XXX ocrd-ocropy-segment throws an exception for buerger_gedichte_1778.ocrd
  #ocrd workspace validate mets.xml

  ocrd-tesserocr-segment-region -l $LOG_LEVEL \
    -m mets.xml -I OCR-D-IMG-BIN -O OCR-D-SEG-REGION
  #ocrd workspace validate mets.xml
  ocrd-tesserocr-segment-line -l $LOG_LEVEL \
    -m mets.xml -I OCR-D-SEG-REGION -O OCR-D-SEG-LINE
  #ocrd workspace validate mets.xml
  # XXX compare ocrd-tesserocr-segment* vs tesseract native

  # XXX This leaves copies of the images at the top level of the workspace, because it "downloads" the "remote" files.
  #     Clean it up. (Maybe only affects ocrd-ocropy-segment)
  find . -maxdepth 1 -name "OCR-D-IMG*" -type f -exec rm -v {} \;
}

do_ocr() {
  # Perform OCR on the segmented lines

  ocrd_tesserocr_recognize_parameters='{ "model": "eng" }'  # TODO mods:language + fontident → model
  remove_filegrp OCR-D-OCR-TESS mets.xml
  ocrd-tesserocr-recognize -l $LOG_LEVEL \
    -m mets.xml -I OCR-D-SEG-LINE -O OCR-D-OCR-TESS \
    -p <(echo $ocrd_tesserocr_recognize_parameters)
  #ocrd workspace validate mets.xml
  # XXX <error>INCONSISTENCY in TextRegion ID 'dummy'
  #     (The whitespace is different)
}

page_fix_xml() {
  # Fix the PAGE XML generated by OCR-D core
  #
  # XXX core does not produce valid XML (See https://github.com/OCR-D/core/issues/242), fix it by setting the correct
  # PAGE XML version. This makes PAGE Viewer open the file.

  filegrp=$1

  local file
  for file in `ocrd workspace find -G $filegrp`; do
    sed -i 's#pagecontent/2017-07-15#pagecontent/2018-07-15#g' $file
  done
}

page_validate_xml() {
  # Validate all PAGE XML against the XML schema

  filegrp=$1

  local file
  for file in `ocrd workspace find -G $filegrp`; do
    xmllint --noout --schema `dirname $0`/xsd/pagecontent.2018-07-15.xsd $file
  done
}

page_fix_image_references() {
  # Make image references relative to the PAGE XML file
  #
  # The rest of OCR-D probably isn't going to like it, but it is a. correct and b. makes PAGE Viewer open the image file
  # automatically.

  filegrp=$1

  local file
  for file in `ocrd workspace find -G $filegrp`; do
    sed -i 's#imageFilename="OCR-D-IMG#imageFilename="../OCR-D-IMG#g' $file
  done
}


do_fontident

do_binarization
do_linesegmentation
do_ocr
page_fix_xml              OCR-D-OCR-TESS
page_validate_xml         OCR-D-OCR-TESS
page_fix_image_references OCR-D-OCR-TESS


# XXX Multiple calls create multiple identical mets:agent elements

# vim:tw=120: