Tesseract OCR

Here is a wrapper script which will convert just about any image to an appropriate input format for tesseract, run tesseract, clean up and dump the output to stdout

imagemagick is required.
Script follows:

#!/bin/bash

tmpid=$$
# echo TID ${tmpid}
convert -compress none -monochrome -resize 200% -density 150x150 $1 /tmp/img.${tmpid}.tif

# cd into the proper directory, since that's where the data dir will be
whichfile=`which $0`
tdir=`dirname $whichfile`
pushd $tdir

./tesseract /tmp/img.${tmpid}.tif /tmp/tout.${tmpid} batch > /dev/null 2>&1
cat /tmp/tout.${tmpid}.txt | perl -e 'while(<>) { $_ =~ s/\|//g; $_ =~ s/\^~R//g; print $_ } '
rm /tmp/tout.${tmpid}.*
rm /tmp/img.${tmpid}.tif
popd

Tis' a sweet script indeed.

Howwwwwwwwwwwwwever,

The convert command leaves something to be desired ... I'm working with jpg images taken straight from a camera - they are slightly blury and have many imperfections - Tesseract is able to parse about 50% of the words, but I'm trying to make it better.

So far things like
convert test.jpg -contrast-stretch 15% test.jpg
convert test.jpg -fill white -tint 70 test.jpg
convert test.jpg -level 20%,80%,1.0 test.jpg
convert test.jpg -sigmoidal-contrast 10,50% test.jpg
convert test.jpg -sharpen 0x2 test.jpg

Followed by conversion to tif have been working for me with limited success.

Does anyone know a set of convert commands the will sharpen text taken from a camera to the point where it can be successfully read by Tesseract?


Thanks,

John

John's changes (which help a ton) added into the wrapper:

#!/bin/bash

# check our args
ARGNO=1 # Number of arguments the script expects.
E_WRONGARGS=65 # It's nice to enumerate the error codes

if [ $# -ne $ARGNO ]
then
echo "Usage: `basename $0` [arg]" >&2 # Error message >stderr.
exit $E_WRONGARGS
fi

if ! [ -f $1 ]
then
echo "File not found: $1"
exit;
fi

# now process and convert our image a bit
tmpid=$$
# echo TID ${tmpid}
TMPNAME=/tmp/tout.${tmpid}.`basename $1 | cut -f2 -d'.'`
cp $1 $TMPNAME
# mogrify ${TMPNAME} -contrast-stretch 15% ${TMPNAME}
convert ${TMPNAME} -density 150x150 -resize 200% -fill white -tint 50 -level 20%,80%,1.0 -sigmoidal-contrast 30,50% -sharpen 0x2 -compress none -monochrome /tmp/img.${tmpid}.tif

# set up our environ for tesseract, which is very picky
# cd into the proper directory, since that's where the data dir will be
whichfile=`which $0`
tdir=`dirname $whichfile`
pushd $tdir > /dev/null

./tesseract /tmp/img.${tmpid}.tif /tmp/tout.${tmpid} batch > /dev/null 2>&1
cat /tmp/tout.${tmpid}.txt | perl -e 'while(<>) { chomp(); $_ =~ s/\|//g; $_ =~ s/\^~R//g; if ($_ =~ /\w/) { print $_."\n" ; }} '
rm /tmp/tout.${tmpid}.*
rm /tmp/img.${tmpid}.tif
popd > /dev/null

How can I install this script on Linux ?

it converts one or many files.
however, did not implement the density and the like ...

****
#!/bin/bash

DELT=TRUE
DELG=TRUE
TARGETDIR=/tmp/
OUTPUT=/tmp/

if [[ -z "$1" ]]
then
echo "Usage: $O [-p] [-t TIF-DIR -o TEXT-DIR -f file ] files_to_OCR
scans all files with tesseract and outputs the text on STOUT

OPTIONS:
-t TIF-DIR saves converted TIF images in target-directory TIF-DIR
-o TEXT-DIR saves files created by tesseract in TEXT-DIR
-f FILE saves all text output in file FILE
-s supress output of text on STOUT (only useful with -t or -o;
do not use with -f)

Examples:
$0 file1 file2 ...
scans the files and displays all the text.

When STOUT is used for the scanned text, progess and error
messages are redirected to tes.error.log. -f redirects output
to file STOUT will show progress and error messages (no log file).
Scanning takes time... be patient :)
"

exit
fi
# redirect error to log.file and fd 5 to stout
# is changed in -f or -s
exec 2>tes.error.log
exec 5>&1


while getopts ":t:o:f:p" Option
do
case $Option in
t ) # save converted tif files in directory
if [[ -d "$OPTARG" ]]
then
TARGETDIR="$OPTARG"/
DELG=FALSE
else
echo "-t: $OPTARG : no such folder"
exit
fi
;;
o ) # save tesseract files in directory
if [[ -d "$OPTARG" ]]
then
OUTPUT="$OPTARG"/
DELT=FALSE
else
echo "-o: $OPTARG : no such folder"
exit
fi
;;
f ) # no STOUT; save in file
FILE="$OPTARG"
exec 5>$FILE
exec 2>&1
;;
s ) # suppress STOUT; show messages
exec 2>&1
exec 5>/dev/null
;;
esac
done


echo "TARGET-DIR=$TARGETDIR" >&2
echo "OUTPUT-DIR=$OUTPUT" >&2


shift $(($OPTIND - 1))


for i in $*
do
if [[ -f "$i" ]]
then
if ! identify "$i" 2>/dev/null 1>&5 # check if graphics file can be converted
then
echo "$i: not convertable" >&2
else
echo "processeing: $i ..." >&2
# converting the graphic file
NEWTIF="$TARGETDIR""${i%\.*}".tif
convert "$i" "$NEWTIF"

# scanning the newly created tif
TEXT="$OUTPUT""${i%\.*}"
tesseract "$NEWTIF" "$TEXT"

# -p: output scanned text on STOUT
# fd #5 is set in getopts
cat "$TEXT".txt >&5

# delete graphic file after use
if [[ $DELG == "TRUE" ]]
then
rm "$NEWTIF"
fi

# delete tesseract output
if [[ $DELT == "TRUE" ]]
then
rm "$TEXT".map "$TEXT".raw "$TEXT".txt
fi

# concatenate all text outputs
if [[ $CAT == "TRUE" ]]
then
cat "$TEXT" >> "$TES"
fi

fi
fi
done

updated script for tesseract
converts and scans one or more files.
***
#!/bin/bash

DELT=TRUE
DELG=TRUE
TARGETDIR=/tmp/
OUTPUT=/tmp/

if [[ -z "$1" ]]
then
echo "Usage: $O [-p] [-t TIF-DIR -o TESS-DIR -f file ] files_to_OCR
scans all files with tesseract and outputs the text on STOUT

OPTIONS:
-t TIF-DIR saves converted TIF images in target-directory TIF-DIR
-o TESS-DIR saves files created by tesseract in TESS-DIR
-f FILE saves all text output in file FILE
-s supress output of text on STOUT (only useful with -t or -o;
do not use with -f)
-c convert with fill white, resize, sigmoidal-contrast, etc..

Examples:
$0 file1 file2 ...
scans the files and displays all the text.

When STOUT is used for the scanned text, progess and error
messages are redirected to tes.error.log. -f redirects output
to file STOUT will show progress and error messages (no log file).
Scanning takes time... be patient :)
"

exit
fi
# redirect error to log.file and fd 5 to stout
# is changed in -f or -s
exec 2>tes.error.log
exec 5>&1


while getopts ":t:o:f:pc" Option
do
case $Option in
t ) # save converted tif files in directory
if [[ -d "$OPTARG" ]]
then
TARGETDIR="$OPTARG"/
DELG=FALSE
else
echo "-t: $OPTARG : no such folder"
exit
fi
;;
o ) # save tesseract files in directory
if [[ -d "$OPTARG" ]]
then
OUTPUT="$OPTARG"/
DELT=FALSE
else
echo "-o: $OPTARG : no such folder"
exit
fi
;;
f ) # no STOUT; save in file
FILE="$OPTARG"
exec 5>$FILE
exec 2>&1
;;
s ) # suppress STOUT; show messages
exec 2>&1
exec 5>/dev/null
;;
c ) # convert picture extensively
CONV=TRUE
;;
esac
done


echo "TIF-output directory=$TARGETDIR" >&2
echo "tesseract output directory=$OUTPUT" >&2


shift $(($OPTIND - 1))


for i in $*
do
I_FILE=${i##.*/} #just the filename w/o directory
if [[ ! -f "$i" ]]
then
echo "$i: file not found" >&2
else
if ! identify "$i" 2>/dev/null 1>&2 # check if graphics file can be converted
then
echo "$i: not convertable" >&2
else
echo "processeing: $i ..." >&2
# converting the graphic file
NEWTIF="$TARGETDIR""${I_FILE%\.*}".tif
#convert "$i" "$NEWTIF"
if [[ $CONV == "TRUE" ]]
then
convert "$i" -density 150x150 -resize 200% -fill white -tint 50 -level 20%,80%,1.0 -sigmoidal-contrast 30,50% -sharpen 0x2 -compress none -monochrome "$NEWTIF" 1>&2
else
convert "$i" -density 150x150 -compress none "$NEWTIF" 1>&2
fi

# scanning the newly created tif
T_FILE="$OUTPUT""${I_FILE%\.*}"
tesseract "$NEWTIF" "$T_FILE" 1>&2

# -p: output scanned text on STOUT
# fd #5 is set in getopts
cat "$T_FILE".txt >&5

# delete graphic file after use
if [[ $DELG == "TRUE" ]]
then
rm "$NEWTIF"
fi

# delete tesseract output
if [[ $DELT == "TRUE" ]]
then
rm "$T_FILE".map "$T_FILE".raw "$T_FILE".txt
fi

# concatenate all text outputs
if [[ $CAT == "TRUE" ]]
then
cat "$T_FILE" >> "$TES"
fi

fi

fi
done

above scripts are buggy do not use
i try to create final usable version

all further versions here. http://www.geocities.com/thierryguy/tess_wrap.html