From 58521d1c5461ab98bc63415ad26e91d54612e43e Mon Sep 17 00:00:00 2001
From: Nick White <git@njw.me.uk>
Date: Wed, 16 Nov 2011 21:46:59 +0000
Subject: Vastly improve ocrpdf script, and create ocrtxt script

---
 extras/mkocrpdf.sh | 36 +++++++++++++++++++++++++++---------
 extras/mkocrtxt.sh | 26 ++++++++++++++++++++++++++
 2 files changed, 53 insertions(+), 9 deletions(-)
 create mode 100644 extras/mkocrtxt.sh

(limited to 'extras')

diff --git a/extras/mkocrpdf.sh b/extras/mkocrpdf.sh
index 3a01559..9113963 100644
--- a/extras/mkocrpdf.sh
+++ b/extras/mkocrpdf.sh
@@ -3,23 +3,41 @@
 #
 # Makes a PDF with embedded text extracted by tesseract
 # Requires imagemagick, pdftk, hocr2pdf and tesseract 3
+#
+# Note: Unfortunately tesseract works much better if one first
+#       makes the image to be OCRed significantly larger. This
+#       script does that, then reduces the results back down
+#       to create a reasonable size PDF.
 
 echo 'tessedit_create_hocr 1' > hocr
 
 for i in `ls *png`
 do
-	a=`basename $i .png`
-	echo processing $a
+	# create a much bigger version of the page image
+	width=`identify "$i" | awk '{print $3}' | sed 's/x.*//'`
+	bigwidth=`expr $width \* 4`
+	convert "$i" -geometry ${bigwidth}x "$i.big.png"
+
+	# scan the page image
+	tesseract "$i.big.png" "$i.big.png" hocr 2>&1 | sed '/Tesseract Open Source OCR Engine/d'
+
+	# this reduces all bbox information to match the original image size
+	sedrule=`cat "$i.big.png.html" \
+	         | sed -e 's/</\n/g' \
+	         | sed -e '/bbox/!d' -e 's/.*bbox//g' -e 's/".*//g' -e "s/'.*//g" \
+	         | awk '{ printf("s/bbox %d %d %d %d/bbox",$1,$2,$3,$4);
+	                  for(i=1;i<5;i++) printf(" %d", $i/4);
+	                  printf("/g\n")}'`
+	sed -e 's/\.big\.png//g' -e "$sedrule" < "$i.big.png.html" > "$i.html"
 
-	# unfortunately tesseract seems to work much better with a
-	# resized larger image
-	convert $i -geometry 1000x $a.big.png
-	tesseract $a.big.png $a hocr 2>/dev/null
+	# combine the image and hocr into a pdf page
+	# Note: hocr2pdf has a habit of segfaulting, so fall back to convert
+	hocr2pdf -i "$i" -o "$i.pdf" < "$i.html" || convert "$i" "$i.pdf"
 
-	# hocr2pdf has a habit of segfaulting, so fall back to convert
-	hocr2pdf -i $a.big.png -o $a.pdf < $a.html || convert $a.big.png $a.pdf
-	rm -f $a.html $a.big.png
+	# remove working files
+	rm -f "$i.big.png" "$i.big.png.html" "$i.html"
 done
 
+# cat the pdf pages together
 pdftk *pdf cat output book.pdf
 rm -f [0-9]*pdf hocr
diff --git a/extras/mkocrtxt.sh b/extras/mkocrtxt.sh
new file mode 100644
index 0000000..02f7146
--- /dev/null
+++ b/extras/mkocrtxt.sh
@@ -0,0 +1,26 @@
+#!/bin/sh
+# See COPYING file for copyright and license details.
+#
+# Makes a text file with text extracted by tesseract
+#
+# Note: Unfortunately tesseract works much better if one first
+#       makes the image to be OCRed significantly larger. This
+#       script therefore temporarily creates a larger file to
+#       feed to tesseract.
+
+for i in `ls *png`
+do
+	# create a much bigger version of the page image
+	width=`identify "$i" | awk '{print $3}' | sed 's/x.*//'`
+	bigwidth=`expr $width \* 4`
+	convert "$i" -geometry ${bigwidth}x "$i.big.png"
+
+	# scan the page image
+	tesseract "$i.big.png" "$i" 2>&1 | sed '/Tesseract Open Source OCR Engine/d'
+
+	# combine the page text with the rest of the book
+	cat "$i.txt" >> book.txt
+
+	# remove working files
+	rm -f "$i.big.png" "$i.txt"
+done
-- 
cgit v1.2.3