A tool for extracting text from scanned documents (via OCR), with user-defined post-processing.

Overview

The project is based on older versions of tesseract and other tools, and is now superseded by another project which allows for more granular control over the text recognition process.

go-ocr

A tool for extracting plain text from scanned documents (pdf or djvu), with user-defined postprocessing.

Motivation

Once I had a task of OCR'ing a number of scanned documents in pdf format. I quickly built a pipeline of the tools to extract images from the input files and to convert them to plain text, but then I realised that modern OCR software is still less than ideal in terms of recognising text, so a good deal of postprocessing was needed in order to remove at least some of those OCR artefacts and irregularities. I ended up with a long pipeline of sed/grep filters which also had to be adjusted per each document and per each document language. What I wanted was a tool that could combine the OCR tools invocation with filters application, also giving an easy way of modifying and combining the filter definitions.

The tool

Given an input file in either pdf or djvu format, the tool performs the following steps:

  1. Images get extracted from the input file using pdfimages or ddjvu tool;
  2. The extracted images get converted to plain text using tesseract tool, in parallel;
  3. The specified filters get applied to the text.

Invocation

go-ocr [OPTION]... FILE

Command line options:

-f,--first N        first page number (optional, default: 1)
-l,--last  N        last page number (optional, default: last page of the document)
-F,--filter FILE    filter specification file name (optional, may be given multiple times)
-L,--language LANG  document language (optional, default: 'eng')
-o,--output FILE    output file name (optional, default: stdout)
-h,--help           display this help and exit
-v,--version        output version information and exit
Example

The following command processes a document some.pdf in Russian, from page 12 to page 26 (inclusive), without any postprocessing, storing the result in the file document.txt:

./go-ocr --first 12 --last 26 --language rus --output document.txt some.pdf

Filter definitions

Filter definition file is a plain text file containing rewriting rules and C-style comments. Each rewriting rule has the following format:

scope type "match" "substitution"

where

  • scope is either line or text;
  • type is either word or regex;
  • match and substitution are Go strings.

Each rule must be on one line.

Each rule of the scope line is applied to each line of the text. There is no processing done to the line by the tool itself other than trimming the trailing whitespace, which means that a line does not have a trailing newline symbol when the rule is applied. After that all the lines get combined into text with newline symbols inserted between them.

Each rule of the scope text is applied to the whole text after all the line rules. All newline symbols are visible to the rule which allows for combining multiple lines into one.

The reason for having two different scopes for the rules is that applying a rule to a line is computationally cheaper that applying to the whole text. Also, this makes the line regular expressions a bit simpler as, for example, \s regex cannot match a newline.

Rules of type word do a simple substitution replacing any match string with its corresponding substitution string.

Rules of type regex search the input for any match of the match regular expression and replace it with the substitution string. The syntax of the regular expression is that of the Go regexp engine. The substuitution string may contain references to the content of capturing groups from the corresponding match regular expression. From the Go documentation, each reference

is denoted by a substring of the form $name or ${name}, where name is a non-empty sequence of letters, digits, and underscores. A purely numeric name like $1 refers to the submatch with the corresponding index; other names refer to capturing parentheses named with the (?P<name>...) syntax. A reference to an out of range or unmatched index or a name that is not present in the regular expression is replaced with an empty slice.

In the $name form, name is taken to be as long as possible: $1x is equivalent to ${1x}, not ${1}x, and, $10 is equivalent to ${10}, not ${1}0.

To insert a literal $ in the output, use $$ in the template.

All filter definition files are always processed in the order in which they are specified on the command line. Within each file, the rules are grouped by the scope, and applied in the order of specification. This allows for each rule to rely on the outcome of all the rules before it.

Rewriting rules examples

Rule to replace ellipsis with a single utf-8 symbol:

line word	"..."  "…"

Rule to replace all whitespace sequences with a single space character:

line regex	`\s+`	" "

Rule to remove all newline characters from the middle of a sentence:

text regex	`([a-z\(\),])\n+([a-z\(\)])` "${1} ${2}"

More examples can be found in the files filter-eng and filter-rus.

In practice, it is often useful to maintain one filter definition file with rules to remove common OCR artefacts, and another file with rules specific to a particular document. In general, it is probably impossible to avoid all manual editing altogether by using this tool, but from my experience, a few hours spent on setting up the appropriate filters for a 700 pages document can dramatically reduce the amount of manual work needed afterwards.

Other tools

Internally the program relies on pdfimages and ddjvu tools for extracting images from the input file, and on tesseract program for the actual OCR'ing. The tool pdfimages is usually a part of poppler-utils package, the tool ddjvu comes from djvulibre-bin package, and tesseract is included in tesseract-ocr package. By default, tesseract comes with the English language support only, other languages should be installed separately, for example, run sudo apt install tesseract-ocr-rus to install the Russian language support. To find out what languages are currently installed type tesseract --list-langs.

Compilation

Invoke make (or make debug) from the directory of the project to compile the code with debug information included, or make release to compile without debug symbols. This creates executable file go-ocr.

Technical details

The tool first runs pdfimages or ddjvu program to extract images to a temporary directory, and then invokes tesseract on each image in parallel to produce lines of plain text. Those lines are then passed through the line filters, if any, then assembled into one text string and passed through text filters, if any. regexp filters are implemented using Regexp.ReplaceAll() function, and word filters are invocations of bytes.Replace() function.

Known issues

Older versions of pdfimages tool do not have -tiff option, resulting in an error.

Platform

Linux (tested on Linux Mint 18 64bit, based on Ubuntu 16.04), will probably work on MacOS as well.

Tools:

$ go version
go version go1.6.2 linux/amd64
$ tesseract --version
tesseract 3.04.01
...
$ pdfimages --version
pdfimages version 0.41.0
...
$ ddjvu --help
DDJVU --- DjVuLibre-3.5.27
...
Lisence: BSD
You might also like...
A set of workflows for corpus building through OCR, post-correction and normalisation
A set of workflows for corpus building through OCR, post-correction and normalisation

PICCL: Philosophical Integrator of Computational and Corpus Libraries PICCL offers a workflow for corpus building and builds on a variety of tools. Th

Toolbox for OCR post-correction

Ochre Ochre is a toolbox for OCR post-correction. Please note that this software is experimental and very much a work in progress! Overview of OCR pos

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)

English | 简体中文 Introduction PaddleOCR aims to create multilingual, awesome, leading, and practical OCR tools that help users train better models and a

Indonesian ID Card OCR using tesseract OCR

KTP OCR Indonesian ID Card OCR using tesseract OCR KTP OCR is python-flask with tesseract web application to convert Indonesian ID Card to text / JSON

Library used to deskew a scanned document
Library used to deskew a scanned document

Deskew //Note: Skew is measured in degrees. Deskewing is a process whereby skew is removed by rotating an image by the same amount as its skew but in

Unofficial implementation of
Unofficial implementation of "TableNet: Deep Learning model for end-to-end Table detection and Tabular data extraction from Scanned Document Images"

TableNet Unofficial implementation of ICDAR 2019 paper : TableNet: Deep Learning model for end-to-end Table detection and Tabular data extraction from

Extract tables from scanned image PDFs using Optical Character Recognition.

ocr-table This project aims to extract tables from scanned image PDFs using Optical Character Recognition. Install Requirements Tesseract OCR sudo apt

Python library to extract tabular data from images and scanned PDFs
Python library to extract tabular data from images and scanned PDFs

Overview ExtractTable - API to extract tabular data from images and scanned PDFs The motivation is to make it easy for developers to extract tabular d

Some bits of javascript to transcribe scanned pages using PageXML

nashi (nasḫī) Some bits of javascript to transcribe scanned pages using PageXML. Both ltr and rtl languages are supported. Try it! But wait, there's m

Comments
  • ocrpdf fails with error message from pdfimages

    ocrpdf fails with error message from pdfimages

    I compiled ocrpdf on linux ubuntu 14.04 but it won't process a pdf file. It issues an error message from pdfimages

    eneafse:~/Downloads$ ocrpdf declasspart4.pdf ERROR: pdfimages version 3.04 Copyright 1996-2014 Glyph & Cog, LLC Usage: pdfimages [options] -f : first page to convert -l : last page to convert -j : write JPEG images as JPEG files -opw : owner password (for encrypted files) -upw : user password (for encrypted files) -q : don't print any messages or errors -cfg : configuration file to use in place of .xpdfrc -v : print copyright and version info -h : print usage information -help : print usage information --help : print usage information -? : print usage information

    Thanks.

    E.J. Neafsey

    opened by Ejneafsey 1
Releases(v0.4.2)
Owner
Maxim
Maxim
かの有名なあの東方二次創作ソング、「bad apple!」のMVをPythonでやってみたって話

bad apple!! 内容 このプログラムは、bad apple!(feat. nomico)のPVをPythonを用いて再現しよう!という内容です。 実はYoutube並びにGithub上に似たようなプログラムがあったしなんならそっちの方が結構良かったりするんですが、一応公開しますw 使い方 こ

赤紫 8 Jan 05, 2023
Train custom VR face tracking parameters

Pal Buddy Guy: The anipal's best friend This is a small script to improve upon the tracking capabilities of the Vive Pro Eye and facial tracker. You c

7 Dec 12, 2021
CNN+LSTM+CTC based OCR implemented using tensorflow.

CNN_LSTM_CTC_Tensorflow CNN+LSTM+CTC based OCR(Optical Character Recognition) implemented using tensorflow. Note: there is No restriction on the numbe

Watson Yang 356 Dec 08, 2022
This is used to convert a string to an Image with Handwritten Characters.

Text-to-Handwriting-using-python This is used to convert a string to an Image with Handwritten Characters. text_to_handwriting(string: str, save_to: s

Akashdeep Mahata 3 Aug 15, 2022
Official code for "Bridging Video-text Retrieval with Multiple Choice Questions", CVPR 2022 (Oral).

Bridging Video-text Retrieval with Multiple Choice Questions, CVPR 2022 (Oral) Paper | Project Page | Pre-trained Model | CLIP-Initialized Pre-trained

Applied Research Center (ARC), Tencent PCG 99 Jan 06, 2023
docstrum

Docstrum Algorithm Getting Started This repo is for developing a Docstrum algorithm presented by O’Gorman (1993). Disclaimer This source code is built

Chulwoo Mike Pack 54 Dec 13, 2022
Scene text recognition

AttentionOCR for Arbitrary-Shaped Scene Text Recognition Introduction This is the ranked No.1 tensorflow based scene text spotting algorithm on ICDAR2

777 Jan 09, 2023
Code for generating synthetic text images as described in "Synthetic Data for Text Localisation in Natural Images", Ankush Gupta, Andrea Vedaldi, Andrew Zisserman, CVPR 2016.

SynthText Code for generating synthetic text images as described in "Synthetic Data for Text Localisation in Natural Images", Ankush Gupta, Andrea Ved

Ankush Gupta 1.8k Dec 28, 2022
Detect textlines in document images

Textline Detection Detect textlines in document images Introduction This tool performs border, region and textline detection from document image data

QURATOR-SPK 70 Jun 30, 2022
chineseocr/table_line 表格线检测模型pytorch版

table_line_pytorch chineseocr/table_detct 表格线检测模型table_line pytorch版 原项目github: https://github.com/chineseocr/table-detect 1、模型转换 下载原项目table_detect模型文

1 Oct 21, 2021
Face Anonymizer - FaceAnonApp v1.0

Face Anonymizer - FaceAnonApp v1.0 Blur faces from image and video files in /data/files folder. Contents Repo of the source files for the FaceAnonApp.

6 Apr 18, 2022
Repository collecting all the submodules for the new PyTorch-based OCR System.

OCRopus3 is being replaced by OCRopus4, which is a rewrite using PyTorch 1.7; release should be soonish. Please check github.com/tmbdev/ocropus for up

NVIDIA Research Projects 138 Dec 09, 2022
Pytorch implementation of PSEnet with Pyramid Attention Network as feature extractor

Scene Text-Spotting based on PSEnet+CRNN Pytorch implementation of an end to end Text-Spotter with a PSEnet text detector and CRNN text recognizer. We

azhar shaikh 62 Oct 10, 2022
A selectional auto-encoder approach for document image binarization

The code of this repository was used for the following publication. If you find this code useful please cite our paper: @article{Gallego2019, title =

Javier Gallego 89 Nov 18, 2022
A real-time dolly zoom camera effect

Dolly-Zoom I've always been amazed by the gradual perspective change of dolly zoom, and I have some experience in python and OpenCV, so I decided to c

Dylan Kai Lau 52 Dec 08, 2022
Generate a list of papers with publicly available source code in the daily arxiv

2021-06-08 paper code optimal network slicing for service-oriented networks with flexible routing and guaranteed e2e latency networkslicing multi-moda

79 Jan 03, 2023
This project modify tensorflow object detection api code to predict oriented bounding boxes. It can be used for scene text detection.

This is an oriented object detector based on tensorflow object detection API. Most of the code is not changed except for those related to the need of

Dafang He 30 Oct 22, 2022
An interactive document scanner built in Python using OpenCV

The scanner takes a poorly scanned image, finds the corners of the document, applies the perspective transformation to get a top-down view of the document, sharpens the image, and applies an adaptive

Kushal Shingote 1 Feb 12, 2022
Introduction to Augmented Reality (AR) with Python 3 and OpenCV 4.2.

Introduction to Augmented Reality (AR) with Python 3 and OpenCV 4.2.

fernanda rodríguez 85 Jan 02, 2023