✨ Use ocrd_olena for binarization

Dockerfile

@@ -15,11 +15,13 @@ RUN apt-get update && \
       python3-pip \
       git \
 # For leptonica/tesseract:
-      cmake libgif-dev libjpeg-dev libpng-dev libtiff-dev zlib1g-dev \
+      cmake libgif-dev libjpeg-dev libpng-dev libtiff-dev zlib1g-dev libpango1.0-dev \
 # For clstm on Ubuntu 19.04:
       swig libeigen3-dev libpng-dev libprotobuf-dev \
 # For cv2:
       libsm6 libxrender1 \
+# For ocrd_olena:
+      wget graphviz imagemagick libmagick++-dev libgraphicsmagick++1-dev libboost-dev \
 # XML utils
       libxml2-utils \
       xmlstarlet \
@@ -47,6 +49,15 @@ RUN curl -sSL -O https://github.com/tesseract-ocr/tessdata_best/archive/$TESSDAT
     rm -rf $TESSDATA_BEST_VERSION.tar.gz
 
 
+# Build ocrd_olena
+RUN curl -sSL -o ocrd_olena.tar.gz https://github.com/OCR-D/ocrd_olena/archive/fde4436.tar.gz && \
+   mkdir ocrd_olena && \
+   tar xvz -C ocrd_olena --strip-components=1 -f ocrd_olena.tar.gz && \
+   cd ocrd_olena && \
+   make install PREFIX=/usr/local && \
+   cd .. && rm -rf ocrd_olena ocrd_olena.tar.gz
+
+
 # Copy OCR models
 RUN mkdir -p /var/lib/calamari-models
 COPY data/calamari-models/GT4HistOCR /var/lib/calamari-models/GT4HistOCR

my_ocrd_workflow

@@ -36,8 +36,9 @@ do_binarization() {
   # Binarize the images
 
   remove_filegrp OCR-D-IMG-BIN mets.xml
-  ocrd-kraken-binarize -l $LOG_LEVEL \
+  ocrd-olena-binarize -l $LOG_LEVEL \
-    -m mets.xml -I OCR-D-IMG -O OCR-D-IMG-BIN
+    -m mets.xml -I OCR-D-IMG -O OCR-D-IMG-BIN \
+    -p <(echo '{"impl": "sauvola-ms-split"}')
 }
 
 do_fontident() {
@@ -143,7 +144,7 @@ page_fix_image_references_to_bin() {
   for file in `ocrd workspace find -G $filegrp`; do
     # Arrays with filenames to the images
     imgs=(`ocrd workspace find -G OCR-D-IMG`)
-    imgs_bin=(`ocrd workspace find -G OCR-D-IMG-BIN`)
+    imgs_bin=(`ocrd workspace find -G OCR-D-IMG-BIN -m image/png`)
 
     # Change all image references to point to the corresponding binarized image
     for i in ${!imgs[@]}; do