OCR — получение текста из изображения с использованием tesseract 3.0 и imagemagick 6.6.5

Я пытаюсь создать сценарий оболочки, который позволяет мне искать текст в изображении. Основываясь на тексте, скрипт сделает все возможное, чтобы получить текст с изображения. Я хотел, чтобы вы внесли свой вклад в это, поскольку этот скрипт, кажется, работает с большинством изображений, но не с теми изображениями, где цвет шрифта текста похож на меньшее окружение вокруг текста.

# !/bin/bash
# 
# imt-ocr.sh is image magick tessearc OCR tool that is used for finding out text in image
#
# Arguments:
# 1     -- image filename (with path)
# 2     -- text to search in image      (default to '')
# 3     -- occurence of text            (default to 1)
# Usage:
# imt-ocr.sh [image_filename] [text_to_search] [occurence]
#

image=$1
txt=$2
occurence=$3    # Default to 1
if [ "$occurence" == "" ]
then
        occurence=1
fi

get_major_color ()
# Returns the major color of an image with its hex value
#       Parameter:      Image filename (with path)
#       Return format:  Returns a string "hex_val_of_color major_color_name"
{
convert $1 -format %c histogram:info: > x.txt
cat x.txt | awk '{print $1}' > x1.txt
h=$(sort -n x1.txt | tail -1);
color_info=$(cat x.txt | grep "$h" | cut -d '#' -f2)
rm -rf x.txt x1.txt
echo "$color_info"
}


invert_color()
# Inverts the color hex value
#       Parameter:      Hex value to be inverted
#       Return format:  Returns in hex
{
input_color_hex=$1                                              # Input color's hex value
white_color_hex=FFFFFF                                          # White color's  hex vlaue
inv_color_hex=`echo $(printf '%06X\n' $((0x$white_color_hex - 0x$input_color_hex)))`
echo $inv_color_hex
}


start_scale=100
end_scale=300
increment_scale=100
tmp_img=dst.tif
attempt=1
for ((scale=$start_scale, attempt=$attempt; scale <= $end_scale ; scale=scale+$increment_scale, attempt++))
        do
                echo "IMT-OCR-LOG: Scaling image to $scale% in attempt #$attempt"
                convert $image -type Grayscale -scale $scale% $tmp_img
                tesseract $tmp_img OUT
                found_oc=$(grep -o "$txt" OUT.txt | wc -l)
                echo "IMT-OCR-LOG: Found $found_oc occurence(s) of text '$txt' in attempt #$attempt"
                if [ $occurence -le $found_oc ] && [ $found_oc -ne 0 ]
                then
                        echo "IMT-OCR-LOG: Printing out the last text found on image"
                        echo "IMT-OCR-LOG: ======================================================"
                        cat OUT.txt
                        echo "IMT-OCR-LOG: ======================================================"
                        rm -rf $tmp_img OUT.txt
                        exit 1
                else
                        echo "IMT-OCR-LOG: Getting major color of image in attempt #$attempt"
                        color_info=`get_major_color $image`
                        true_color=$(echo $color_info | awk '{print $2}')
                        true_val=$(echo $color_info | awk '{print $1}')
                        echo "IMT-OCR-LOG: Major color of image is '$true_color' with hex value of $true_val in attempt #$attempt"

                        # Blur the image
                        echo "IMT-OCR-LOG: Bluring image in attempt #$attempt"
                        convert $tmp_img -blur 1x65535 $tmp_img

                        # Flip the color
                        inverted_val=`invert_color $true_val`
                        echo "IMT-OCR-LOG: Inverting the major color of image from 0x$true_val to 0x$inverted_val in attempt #$attempt"
                        convert $tmp_img -fill \#$inverted_val -opaque \#$true_val $tmp_img

                        # Sharpen the image
                        echo "IMT-OCR-LOG: Sharpening image in attempt #$attempt"
                        convert $tmp_img -sharpen 1x65535 $tmp_img

                        # Find text
                        tesseract $tmp_img OUT
                        found_oc=$(grep -o "$txt" OUT.txt | wc -l)
                        echo "IMT-OCR-LOG: Found $found_oc occurence(s) of text '$txt' in attempt #$attempt"
                        if [ "$found_oc" != "0" ]
                        then
                                if [ $occurence -le $found_oc ]
                                then
                                        echo "IMT-OCR-LOG: Printing out the last text found on image"
                                        echo "IMT-OCR-LOG: ======================================================"
                                        cat OUT.txt
                                        echo "IMT-OCR-LOG: ======================================================"
                                        rm -rf $tmp_img OUT.txt
                                        exit 1
                                fi
                        fi
                fi

                rm -rf OUT.txt

        done

rm -rf $tmp_img

Вот пример с проблемой, изображение (test.jpeg) http://www.igoipad.com/wp-content/uploads/2012/07/03-Word-Collage-iPad.jpeg

[admin@ba-callgen image-magick-tesseract-processing]$ sh imt-ocr.sh test.jpeg Common
IMT-OCR-LOG: Scaling image to 100% in attempt #1
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Common' in attempt #1
IMT-OCR-LOG: Getting major color of image in attempt #1
IMT-OCR-LOG: Major color of image is 'grey96' with hex value of F5F5F5 in attempt #1
IMT-OCR-LOG: Bluring image in attempt #1
IMT-OCR-LOG: Inverting the major color of image from 0xF5F5F5 to 0x0A0A0A in attempt #1
IMT-OCR-LOG: Sharpening image in attempt #1
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Common' in attempt #1
IMT-OCR-LOG: Scaling image to 200% in attempt #2
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 1 occurence(s) of text 'Common' in attempt #2
IMT-OCR-LOG: Printing out the last text found on image
IMT-OCR-LOG: ======================================================
Settings M...
Text
Common words
Exclude numbers
word case
Theme & Layuul
Color theme
Fnnl
Word layout
Clrien lalion
7301
Lrmclsc ape
\u2018OTC
Ergl sw v.-ords >
li( `
I):Jntc1'\:1r\qa )
Landon Spring >
Hough Trad >
H3'fJ|1d :-Ialf >
H L

IMT-OCR-LOG: ======================================================
[admin@ba-callgen image-magick-tesseract-processing]$ 
[admin@ba-callgen image-magick-tesseract-processing]$ 
[admin@ba-callgen image-magick-tesseract-processing]$ 
[admin@ba-callgen image-magick-tesseract-processing]$ 
[admin@ba-callgen image-magick-tesseract-processing]$ 
[admin@ba-callgen image-magick-tesseract-processing]$ 
[admin@ba-callgen image-magick-tesseract-processing]$ sh imt-ocr.sh test.jpeg Portrait
IMT-OCR-LOG: Scaling image to 100% in attempt #1
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Portrait' in attempt #1
IMT-OCR-LOG: Getting major color of image in attempt #1
IMT-OCR-LOG: Major color of image is 'grey96' with hex value of F5F5F5 in attempt #1
IMT-OCR-LOG: Bluring image in attempt #1
IMT-OCR-LOG: Inverting the major color of image from 0xF5F5F5 to 0x0A0A0A in attempt #1
IMT-OCR-LOG: Sharpening image in attempt #1
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Portrait' in attempt #1
IMT-OCR-LOG: Scaling image to 200% in attempt #2
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Portrait' in attempt #2
IMT-OCR-LOG: Getting major color of image in attempt #2
IMT-OCR-LOG: Major color of image is 'grey96' with hex value of F5F5F5 in attempt #2
IMT-OCR-LOG: Bluring image in attempt #2
IMT-OCR-LOG: Inverting the major color of image from 0xF5F5F5 to 0x0A0A0A in attempt #2
IMT-OCR-LOG: Sharpening image in attempt #2
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Portrait' in attempt #2
IMT-OCR-LOG: Scaling image to 300% in attempt #3
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Portrait' in attempt #3
IMT-OCR-LOG: Getting major color of image in attempt #3
IMT-OCR-LOG: Major color of image is 'grey96' with hex value of F5F5F5 in attempt #3
IMT-OCR-LOG: Bluring image in attempt #3
IMT-OCR-LOG: Inverting the major color of image from 0xF5F5F5 to 0x0A0A0A in attempt #3
IMT-OCR-LOG: Sharpening image in attempt #3
Tesseract Open Source OCR Engine with Leptonica
IMT-OCR-LOG: Found 0 occurence(s) of text 'Portrait' in attempt #3
[admin@ba-callgen image-magick-tesseract-processing]$ 

Как видите, я могу найти текст "обычный", но не "портрет". Причина в цвете шрифта портрета. Любая помощь в улучшении этого скрипта...

Я использую Центос 5.