Bratislava WS - Fuchs - Abbyy - OCR overview_pdf

IMPACT is supported by the European Community under the FP7 ICT Work Programme. The project is coordinated by the National Library of the Netherlands.

Michael FuchsSenior Product Marketing ManagerABBYY Europe

[email protected]

Optical Character Recognition (OCR) Introduction & Overview


IMPACT + ABBYY - OCR Introduction & Overview 2

Agenda ABBYY Technology in the IMPACT project

Who is ABBYY? Company Overview Product Overview How is OCR used in real-life scenarios?

Optical Character Recognition - Basics What is OCR? How does OCR work inside? OCR = Only Character Recognition? IMPACT – the areas of improvement

Questions & Answers


IMPACT & ABBYY




Improving Access to Text Mission of IMPACT: It aims to significantly improve access to historical text and

remove the barriers that stand in the way of the mass digitisation of the European cultural heritage.

Partners: Koninklijke Bibliotheek, The British Library, Österreichische Nationalbibliothek, Universität Innsbruck,

Deutsche Nationalbibliothek, Bayerische Staatsbibliothek, Staats- und Universitätsbibliothek Göttingen

ABBYY, IBM Israel – Science and Technology Ltd, Instituut voor Nederlandse Lexicologie

National Centre for Scientific Research "Demokritos“,

Centrum für Informations- und Sprachverarbeitung, University of Munich

University of Bath, University of Salford, Bibliothèque Nationale de France

Web: www.impact-project.eu


IMPACT & ABBYY

ABBYY is the OCR technology provider for IMPACT members

IMPACT members work with ABBYYs OCR SDK (FineReader Engine), because: Only development toolkits allow developers to combine new/different modules,

for example: complex dictionaries Scientific research & tests have to be implemented in custom modules



IMPACT & ABBYY

ABBYY improves the OCR core technologies for the recognition of old documents, current focus areas are Image pre-processing Character recognition

IMPACT currently focuses on research and not in setting up a production system ;o)

Improvements in ABBYY recognition technologies that are a result of the IMPACT project will be added to future products Important: ABBYY FineReader 8/9/10 Professional (Box) has NO Fraktur OCR Fraktur OCR is only available in Recognition Server und FineReader Engine


IMPACT + ABBYY – OCR Introduction & Overview

ABBYY – an Overview

ABBYY & OCR for IMPACT

Who is ABBYY?

Leading developer of artificial intelligence software in document recognition, data capture and linguistics

Headquartered in Moscow, Russia

Founded in 1989 by Mr. David Yang as BIT Software

More than 880 employees worldwide

8 offices worldwide

Established sales and distribution network in more than 130 countries worldwide


ABBYY Worldwide

ABBYY JapanFremontABBYY USA ABBYY Ukraine

Kiev

ABBYY Europe UK

ABBYY Europe GmbH Munich, Germany

ABBYY Taiwan

ABBYY Headquarters/ ABBYY RussiaMoscow


ABBYY in Western Europe

ABBYY Europe GmbH

Located in Munich, Germany

Established in 2001

Serves partners and customers in Western European countries

Sales and Marketing

Sales● Distribution, channel development, partner management

Marketing● Product marketing, channel marketing, outbound marketing (PR, advertising, direct)

More than 50 employees today


Product Overview


ABBYY Product Brands Mainline Distribution

“Box” products:

ABBYY FineReaderOptical character recognition (OCR)/text processing end user products

ABBYY FotoReaderConversion of texts taken with digital cameras

ABBYY PDF TransformerPDF conversion and creation for end users

ABBYY LingvoElectronic dictionaries, Russian and European languages


ABBYY Product Brands Direct Sales and VAR Distribution

Licensing and integration products:

ABBYY Recognition ServerServer-based OCR

ABBYY FormReader and ABBYY FlexiCaptureForm processing, unstructured document processing, document assembly

ABBYY FineReader Engine SDKComprehensive toolkit for integrating recognition and data capture technologies into third-party applications

ABBYY Mobile OCR EngineOCR for thin clients such as mobile phones, PDAs and Web applications


ABBYY OCR Products – Usage View

Desktop/Workgroup Server/Backend SDK/Integration

OCR

& D

ocum

ent

Conv

ersi

on FineReader (Professional, Corporate, Site Licence Edition)

PDF Transformer

FotoReader

ScreenshotReader

Recognition Server (Professional, Extended Edition)

FineReader Engines (Windows, Linux, Mac OS X, Free BSD, Embedded Systems)

Mobile OCR Engine (Android, Symbian, Linux, Windows, Windows Mobile,iPhone )

End Users, Companies,(Libraries)

Companies,Scan Service Provider,

Libraries

User driven processing,Ready to use

Automated processing,Ready to use

Automated processing,Development needed

Developers,Scan Service Provider

IMPACT ResearchUser

sar

e:


OCR Basics

ABBYY & OCR for IMPACT 10

Designed to be not OCRed

ABBYY & OCR for IMPACT 11

What (ABBYY) OCR can read...

Recognition Languages >191 languages altogether Alphabets: Cyrillic, Latin, Greek, Armenian, Hebrew, Thai 34 languages with dictionary support and spell check Chinese, Japanese, Korean (CJK) - 4 sets of hieroglyphs

(Chinese (traditional and simplified), Japanese, Korean) 5 languages in FineReader XIX (Gothic and other 17-20 century fonts) 6 programming languages (Basic, C/C++, COBOL, Java, etc.) 4 artificial languages (Esperanto, Interlingua, etc.) Simple chemical formulas

Font Types Recognition of mixed font types

(dot-matrix printer, typewriter, Gothic, etc.) OCR-A OCR-B MICR (E13B) CMC-7


Step 1. Scanning, Image Loading, Pre-Processing and Modification Compensating image defects and making the document better viewable and suited for

automatic OCR

Step 2. Document Layout Analysis Detect sections of a document, analyze layout and find barcodes

Step 3. Character Recognition Automatic recognition of characters, apply selected recognition languages, dictionaries

and other settings

Step 4. Verification by Operators (optional) Manual validation of suspicious characters and words

Step 5. Document Synthesis and Export Generating an output document in the selected format

12

OCR Processing Steps


Step 1. Image Loading, Pre-Processing and ModificationImages from existing files or captured with a scanner

Splitting images

Scaling (e.g. low resolution images can be digitally magnified)

Rotation (on 90, 180, or 270 degrees)

Flipping and inverting images

Cropping (selecting rectangular areas)

Creating previews (small images for previews)

Changing text colour and background in rectangular areas

13



Step 1. Image Loading, Pre-Processing and ModificationCompensating for scanning defects

Automatic de-skew to proper straight position

Straightening text lines Controlled de-speckle

(cleaning garbage dots)

14

ABBYY OCR Processing Steps


Step 1. Image Loading, Pre-Processing and Modification

Intelligent background filtering

Adaptive Binarisation

15


General binarisation on an image level can not deliver good results for OCR


Step 1. Image Loading, Pre-Processing and Modification

Success during IMPACT

16


Original State of Art New

No text from the other page


New Binarization Examples

23

Original scan

Prev. binarization

New binarization


Camera OCRAutomatic correction of 3D perspective distortions

24

Before

After


Camera OCRISO noise reduction

25

Before

After


Step 2. Document Layout AnalysisDetecting sections of a document, analyze layout and find barcodes

20



Step 3. Character RecognitionAfter line detection, character recognition is applied with different classifiers

21


Raster classifier Contour classifier

Feature differentiating classifier Structure classifier


Step 3. Character Recognition – learn new symbolsOwn Pattern Training to learn special characters on a pixel level

22

OCR Optimization


Step 3. Character Recognition – back to the word levelApplying selected recognition languages and dictionaries

Own languages and dictionaries can be defined

23

OCR Optimization


Step 4. Verification by Operators (optional)Manual validation or correction of Layout Analysis Results

● Text blocks● Image blocks● Table blocks

Suspicious characters and wordcorrections using dictionaries

Re-Recognition with other language settings

Recognition Server allows one to set quality level and also to log processing results in a XML file

24



Step 5. Document Synthesis and ExportGenerating an output document in the selected format

TXT, Office formats, PDF, etc.

From version 9.0 on ADRT(Adaptive DocumentRecognition Technology) included.Goal: Understanding the document structure and detectinge.g. headers, footers, footnotes. V10: table of contents

SDKs and Recognition Server offer more export formats, e.g.● XML● Internal

FineReader Engine Format

25

ABBYY OCR Processing Steps


OCR in General&

IMPACT in Particular


Recreates the same layout as in the original document Resulting document looks just like the scanned original Information captured during Layout Analysis is used here

Supports popular document formats ABBYY products support all popular output formats the customer needs

PDF, PDF/A, XML, HTML, TXT/CSV, Word, Excel, PowerPoint and DBF

Supports image output BMP, PCX, JPEG, JPEG 2000, TIFF, PNG

Compliance with the regulations Support for selective access password protection, document encryption,

support for PDF/A format, etc.

27

OCR = Only Character Recognition?


Step 1. Image Quality Problem areas: Scans of microfilms, distortions, shine through characters Optimisation approach: Image pre-processing, e.g: Binarisation

Step 2. Document Analysis Problem areas : Layout of old print material, e.g. narrow columns in old newspapers, Optimisation approach: improved Layout/Document Analysis

Step 3. Character recognition & Languages Problem areas : Used Fonts, old language (grammar & spelling) Optimisation approach: Optimised patterns, adaptive OCR, creation of special dictionaries

Step 4. Validation & Correction Problem areas : often recurring errors during Fraktur OCR, Scalability of correction Optimisation approach: New approaches for mass verification

Step 5. Document Synthesises, Export & Rating Problem areas : Content classification, Meta data generation, “reliable ”formats Optimisation approach: XML, AltoXML, XML analysis, PDF/A, …

28

IMPACT = „Step by Step“ Optimisation


Thank you for your attention!

Questions?

Michael FuchsSenior Product Marketing ManagerABBYY [email protected]

Bratislava WS - Fuchs - Abbyy - OCR overview_pdf

Technology

scan service

optical character

ocr processing

ocr introduction

impact abbyy

character

optimisation

introduction