cli tool for inline pdf ocr on mac osx

ocr.sh

#!/bin/bash
# ocr.sh for Mac OSX 10.11
# works in memory (~/.ocr-TIMESTAMP)
# needs pdftk, pdftoppm, tesseract-ocr
# Port from multithreaded OCR with tesseract 3.03/3.04 (Michael Luthardt <edv@dr-luthardt.de> 2016, https://dr-luthardt.de/linux.htm?tip=pdfx)

# PREREQUISITES (requires homebrew)
# 1 - tesseract: brew install tesseract --with-all-languages (all languages will take up to 1.2gb space, see https://github.com/tesseract-ocr/tesseract/wiki to install ony selective)
# 2 - pdftk: https://www.pdflabs.com/tools/pdftk-the-pdf-toolkit/pdftk_server-2.02-mac_osx-10.11-setup.pkg
# 3 - pdftotext, pdftoppm: brew install poppler

WORKDIR="$HOME/.ocr-$(date +"%m_%d_%Y")"

# check for valid PDF as first argument
if (! [ $# -ge 1 ] || ! (gs -q -o nul -sDEVICE=nullpage -dFirstPage=1 -dLastPage=1 "$1" &>/dev/null)); then
	cat << eot
	First argument is not a valid PDF or doesn't exist.
	Usage: [path/to/]pdffile [lan] [-y]
		See tesseract --list-langs for installed languages.
eot
	exit 1
fi

# check for parameters
# language as 2nd argument, if not, set default (eng+deu)
# it's your own task to provide lan.traineddata
# with option -y OCR is forced anyway

FORCE=0 # as a precaution
if [ ${#} -eq 1 ]; then
	LANG=eng+deu
else
	[ ${#2} -eq 3 ] && LANG=$2 || LANG=eng+deu
	[ "x$2" == "x-y" ] && FORCE=1
	[ "x$3" == "x-y" ] && FORCE=1
fi

if [ $FORCE -eq 0 ]; then
#check if PDF already contains a text layer
	unset -v ANS	# as a precaution
	if [ $(pdftotext "$1" - | grep -cE '[[:alpha:]]' 2>/dev/null) -ne 0 ]; then
		echo -e "\nThis PDF contains a text layer."
		read -r -p "Proceed anyway? Old text will be removed. [N|y] " ANS
		[ "x$ANS" = "xy" ] || { echo; exit 2; }
	fi
fi

FILE=`basename "$1"`

# find number of cpus, calculate maxjobs
NCPUS=$(sysctl -n hw.ncpu)
MAXJOBS=$((($NCPUS+1)/2))

# work in memory:
mkdir -p $WORKDIR
cp "$1" $WORKDIR/
cd $WORKDIR
rm -f pg_* pg-* # as a precaution

# burst input PDF into pages
tput bold; echo -e "\n${FILE}:"; tput sgr0
pdftk "$FILE" burst
echo "`sed -n '/NumberOfPages/s/NumberOfPages: //p' doc_data.txt` pages to process ..."

# bundle all actions into one function
echo 0 > err
ocr() 
{
    pdftoppm -thinlinemode shape -singlefile -png -r 200 $PAGE ${PAGE%.*} &>/dev/null
    echo $? > err
    tesseract -l $LANG ${PAGE%.*}.png ${PAGE%.*} pdf &>/dev/null
    ERR=$?; echo $((`cat err` + $ERR)) > err
    [ `cat err` -eq 0 ] && { tput civis; echo -en "        ... $PAGE done\r"; tput cnorm; } \
    || { tput civis; echo -en "\n        ... $PAGE something went wrong\n"; tput cnorm; }
}

# do orc() in background
for PAGE in pg_*.pdf; do
 	[ `cat err` -ne 0 ] && break
	ocr $PAGE &
	# but limit the number of simultaneous jobs
 	[ `jobs -p | wc -l` -ge $MAXJOBS ] && wait
done

# wait for last bg job to finish – important!
wait

# leave memory
cd - &>/dev/null

# concatenate pdf's to input_ocr.pdf
if [ `cat $WORKDIR/err` -eq 0 ]; then
	pdftk $WORKDIR/pg_*.pdf cat output "${1%.*}_ocr.pdf"
	[ $? -eq 0 ] && echo -e "\n\n\t... ${1%.*}_ocr.pdf created\n" \
	|| echo -e "\n\n\t... Error encountered.  No output created.\n"
else
	echo -e "\n\\t... Errors encountered.  No output created.\n"
fi

# clear memory
rm -rf $WORKDIR

exit

batch use:

Copy ocr.sh in an executable path without .sh file ending. Run ls *.pdf | xargs -I '{}' ocr {}.