Automatically Rotating Scanned Text Images with Tesseract OCR

Problem

If you've ever had a batch of scanned images with text, you know how tedious it can be to manually rotate each one to the correct orientation. This process can be especially frustrating when dealing with a large number of images. Wouldn't it be great if there were a way to automatically rotate these images so that the text is always upright and readable?

Solution

To solve this problem, I developed a simple script that automatically detects the correct orientation of text in scanned images using Optical Character Recognition (OCR) and dictionary matching. Here's how it works:

  1. OCR Parsing with Tesseract: I used Tesseract, a popular open-source OCR engine, to extract text from the images. Tesseract is powerful and versatile, making it an excellent choice for this task.

  2. Dictionary Matching: I created a list of the most commonly occurring words in the text. This list acts as a reference to determine the correct orientation. While my example includes only 5-6 words, you can easily expand this list to improve accuracy.

  3. Rotation and Validation: The script rotates each image in increments (90°, 180°, 270°) and re-runs the OCR. It then compares the recognized text against the dictionary. Given that OCR is not 100% accurate, I accounted for minor deviations when matching the words. The image is rotated to the orientation where the most dictionary words are recognized, ensuring the text is upright.

How to Use

To get started, simply copy the three files listed below into your ~/bin directory and run the tesseract_rotate_all script. The script will automatically process all images in the directory, rotating them to the correct orientation based on the detected text.

Dependencies

Before running the script, you need to install the perl re::engine::TRE - a regular expression engine that handles approximate matching, which is essential for dealing with OCR inaccuracies.

recognize_good_rotation

       


#use lib "/root/perl5/lib";
# @author Miroslav Bodis 2014

use strict; 
use warnings;

my $file = shift;
 my $rotati
my $debug_mode = shift;

my $find = 0;

if (!defined $debug_mode){
 $debug_mode = 0; 
 # $debug_mode = 1; # TODO use for more details 
}

if ($debug_mode == 1){ 
 print "file:" . $file."\n";
 print $rotation . "\n";
 print "debug mode: " . $debug_mode."\n";
}

my @recognize_words = ('then', 'change', 'when', 'over', 'suddenly', 'another');

open(my $fh, "<", $file) or die "cannot open file";

while(<$fh>)  {
 chomp;
 
 my $line = $_;
 {
  use re::engine::TRE max_cost => 1;

  foreach (@recognize_words) {

   if ($line =~ /$_/i) {
    
    $find += 1;

    if ($debug_mode == 1){ 
     print "match word: " . $_ . "\n";
    }    
   }
  }
 }
}

close $fh;

if ($find > 2){
 exit 0;
}

exit 1;



tesseract_rotate

       

#! /bin/sh

# @author Miroslav Bodis 2014


if [ -z "$1" ]
then
 echo "
 @author Miroslav Bodis 2014
 
 #   script to autorotate image of printed text
 # - inpnut image try all 4 rotations (0, 90, 270, 180)
 # - tesseract current rotation
 # - use your dictionary to find word in tesseracted text (with some tollerance - used TRE max_cost => 1)
 #   - see log for results
 #   - TODO: copy script to your bin folder e.g.: \"~/bin/recognize_good_rotation.pl\"
 # \$1 -> \"input_image\"" 
 exit
fi;


# required 1 arguments
if [ -z "$1" ]
then
 echo "required 1 arguments \"image_name\""
 exit
fi;


help_rotated_img="rotation_help.jpg"
help_ocr_out="output_ocr"
help_ocr_out_txt="$help_ocr_out.txt"
find=1


# 0 - rotation
echo "image $1 try rotation 0"
tesseract -l slk $1 $help_ocr_out 
perl ~/bin/recognize_good_rotation.pl $help_ocr_out_txt 'rotation 0'
find=$?


# 90 - rotation clockwise
if [ $find -eq 1 ] 
then

 echo "image $1 try rotation 90"
 convert $1 -rotate 90 -quality 100 $help_rotated_img
 tesseract -l slk $help_rotated_img $help_ocr_out
 perl ~/bin/recognize_good_rotation.pl $help_ocr_out_txt 'rotation 90'
 
 find=$?
fi;


# 270 - rotation clockwise
if [ $find -eq 1 ] 
then
 
 echo "image $1 try rotation 270"
 convert $1 -rotate 270 -quality 100 $help_rotated_img 
 tesseract -l slk $help_rotated_img $help_ocr_out 
 perl ~/bin/recognize_good_rotation.pl $help_ocr_out_txt 'rotation 270' 

 find=$?
fi;

# 180 - rotation clockwise
if [ $find -eq 1 ] 
then
 
 echo "image $1 try rotation 180"
 convert $1 -rotate 180 -quality 100 $help_rotated_img
 tesseract -l slk $help_rotated_img $help_ocr_out
 perl ~/bin/recognize_good_rotation.pl $help_ocr_out_txt 'rotaiton 180'

 find=$? 
fi;


if [ $find -eq 1 ] 
then
 echo ">>>>>>>>>>>>>>>>> image $1 NOT ROTATED, please update dictionary ! <<<<<<<<<<<<<<<<"
else 
 echo "image $1 ROTATED"
 # if rotated replace new right rotation with old one
 cp $help_rotated_img $1
fi;

rm $help_rotated_img
rm $help_ocr_out_txt

tesseract_rotate_all

       

#! /bin/sh

# @author Miroslav Bodis 2014
# move to current folder with pictures and run "tesseract_rotate_all"

FILES=./*

for f in $FILES
do 
 echo "--- --- --- --- START FILE $f" 
 tesseract_rotate $f
done

echo "finished"

Comments

Post a Comment

Popular posts from this blog

Skate Tricks Recognition Using Gyroscope

Image
In this article, we’ll describe how to recognize skateboard tricks using a gyroscope. This sensor is already present in most smartphones, but if you’re not familiar with it, here is a description . Before we dive in, a small disclaimer: This project originated as part of a Slido hackathon with the goal of demonstrating that it’s possible to recognize skateboard tricks using a gyroscope. For simplicity, we’ll focus on recognizing just two basic tricks and won’t consider the skater’s stance on the board. We’ll break down the problem into several smaller tasks: recording the trick, storing the data, describing the trick, analyzing it, and, finally, recognizing it. Recording the Trick Let’s start with trick recording. To obtain precise data, the device with the gyroscope needs to be attached directly to the skateboard. Place it on the bottom of the board to avoid interfering with the execution of the trick. The device should also be compact and lightweight to avoid altering the board’s cen...
Read more

Play table

Image
Using proximity sensors for playing midi tones combined with LED visualization Description The goal of this project was to create a table-sized device equipped with multiple proximity sensors capable of playing MIDI tones. Each sensor is accompanied by LEDs that indicate the distance of the user's hand above the table. This interactive table can be used by one or more people simultaneously. Hardware Setup I began with a cardboard prototype to test the sensors and the underlying logic. Afterward, I ordered a custom plottered sticker with a design which was painted with bare conductive paint. I drilled holes and connected the touchboard to seven Arduino Nanos, each controlling 13 LEDs. Additionally, I incorporated two potentiometers: one for volume control and another for changing the note setup. Programming The programming task was divided into two parts: Master Program (Touchboard): Reads values from the proximity sensors and sends messages to the slave Arduinos. Sl...
Read more

Counting dice and train wagons using computer vision

Image
Computer vision exercises with preprocessing Before starting my next project, I decided to work on some computer vision exercises. Each example is based on straightforward image preprocessing techniques. No complex data structures or machine learning are involved. Dice Detection I got this idea while browsing the net and became curious about how challenging it would be to write such a script. Here’s a step-by-step breakdown of the algorithm: Movement Detection : By comparing several frames with thresholds, we can determine if there is any movement in the frame. Adding a small time buffer after the movement stops gives us more accurate information. Removing the Background : Thresholding the grayscale frame helps to remove the background, leaving us with a binary image that highlights the objects. Cropping the Objects : Using contours, we can detect and isolate the objects by cropping them. Detecting Dots : By inverting the image, the dots on the dice become more distinguish...
Read more