The CSDK16 Development Environment
Whats Improved in OmniPage Capture SDK 16

IMPROVED!

Superior Accuracy and Layout Retention
Word accuracy is up to 34% greater than alternatives on Windows* Word accuracy is up to 62% greater than alternatives on Linux* Layout retention to 27% improvement over v15
Fastest Multi-Page OCR with Multi-Core Processors
Can speed up processing up to 46% over v15 on dual-core machines Can speed up processing by 62% over v15 on quad-core machines
IMPROVED! IMPROVED! IMPROVED!
Superior Text PDF Parsing Accuracy Common Western and Asian API for C/C++ Code samples and watchdog utility for reliable 24/7 applications
*Actual results may vary based on machine, input, output and implementation 2
Whats New in OmniPage Capture SDK 16

Windows Features

NEW! True Open Office XML Output
Office 2007 XPS Image, Normal (Text), Searchable

NEW! NEW! NEW! NEW!

Recognizes Mixed Asian & Western Pages within Documents Electronic/Paper Forms Data Collection Document Redaction and Highlighting 3DC Digital Camera/Image Correction
Linux & Macintosh 15.5
Subset of CSDK Windows Features
Same API as Windows Kit for Easy Development Western and Asian Image Preprocessing 2-way voting for Superior Accuracy Text Only (No Layout/RTF) PDF Module for PDF Image and Searchable PDF
Structure of OmniPage Capture SDK v16

Engine Choices

Engines for Different Data Representations
OmniFont text Dot Matrix text 1D and 2D Barcodes Optical Marks (OMR) Handprinted text

Several OmniFont Engines

3 different stand-alone engines: MTX, MOR, FRX 2 different voting engines: PLUS2W, PLUS3W Asian engine (Chinese, Japanese, Korean)

Speed/Accuracy Trade-off

3 Trade-off settings for most OmniFont engines
Word Speed/Accuracy Scalability Windows SDK

OCR Process Scalability

on CSDK 16
4x FRX+MGO+MOR+BR3 P3W_Formatted: MOST ACCURATE DEFAULT 3x Accuracy FRX+MGO+MOR+BR3 P3W: FRX+MGO+BR2 P2W_Formatted: FRX+MGO+BR2 P2W: FRX+Legacy: MTX+FRX-P+MOR+BR3 P3W_Balanced: MTX+FRX-P+MOR_2x+BR3 P3W_Fast: 2x MTX+MOR+BR2 P2W_Balanced: FRX: MTX: MOR: 3x

1x 1x 2x Speed

FASTEST
Hardware Specification for speed benchmarking
Intel Core Duo 3 GHz, 32bit, 2GB RAM

Output Choices

LETTER structure Fast Direct Text Conversion Document Conversion (RecAPIPlus)

LETTER structure

kRecGetLetters() All low-level information is available Character code (in Unicode) Bounding box co-ordinates Zone index, cell index Font attributes bold, italic, underlined serif or sans-serif monospaced or proportional Language
Fast Direct Text Conversion Document Conversion (RecAPIPlus)
LETTER structure Fast Direct Text Conversion
kRecConvert2DTXT() Best for unformatted text output Normal Text CSV records Searchable PDF (Image on Text) XML with coordinates
Document Conversion (RecAPIPlus)
RecCreateDoc(), RecInsertPage(), RecConvert2Doc() Up to 500 pages Both Page and Document Formatting is supported Several document formats: PDF 1.6, PDF/A, PDF-MRC, Searchable PDF XPS: both Normal and Searchable Office 2007 (.docx,.xlsx,.pptx) XML Many others
Unsurpassed PDF functionality
Integrated toolkit allows for PDF content re-purposing Conversion From and To PDF Achieve up to 100% conversion with superior PDF Parsing
PDF Overlay Matching capabilities
Output PDF 1.6 and PDF/A files Generates PDF files up to 84% smaller in size

PDF-MRC High Compression

Create searchable PDF files for archive and document management systems
Output for Office 2007 and XPS
Other OCR SDKs Do Not Create the Proper OOXML Format Save Directly to Native Office 2007 XML Formats
Includes Word (.docx), Excel (.xlsx) & PowerPoint (.pptx) Co-developed and financed by Microsoft Smaller File Size Easier Document Recovery Open Source Format Scan to Image, Searchable and Normal XPS The first Searchable XPS solution Load and convert Image and Normal XPS to any format The first XPS conversion solution Convert PDF Documents to XPS

Full XPS Support

Development Environment Choices
Conventional Interface Kernel API Rec API Plus PInvoke Interface IPRO COM Interface

Conventional Interface Kernel API
C/C++ Page oriented functions The fastest CSDK API LETTER array or Direct Text output Thread safe Scales up to 4 cores E.g. kRecProcessPagesEx()
Rec API Plus PInvoke Interface IPRO COM Interface
Conventional Interface Kernel API Rec API Plus
C/C++ Document oriented functions Wide range of output formats Requires more resources than KernelAPI
PInvoke Interface IPRO COM Interface
Conventional Interface Kernel API Rec API Plus P/Invoke Interface
Best for managed.NET applications C#, Managed C++, VB.NET All the KernelAPI and RecAPIPlus functions are available Add CAPI_PInvoke.dll to your projects References Nuance.OmniPage.CSDK.RecAPI and RecAPIPlus namespaces

IPRO COM Interface

Conventional Interface Kernel API Rec API Plus P/Invoke Interface IPRO COM Interface
Object Oriented Interface In-proc COM server (IProPlus.dll, Visuals.dll) ActiveX Visual Toolbox Controls: Image viewer, Zone viewer, Text editor, and others Use it only if the Visual Toolbox controls are needed Out-of-proc COM server (IproPlusExe.exe) The P/Invoke interface is usually more desirable
Visual Toolbox Components
Allows Developers to Build Quickly
Polished and Sophisticated Applications No Significant Programming

Workflow Assistant

Visually define workflow and track execution

Image display

Image Viewer Control (IVC)

Correction editors

Text Editor Control (TEC)

Results viewing

Thumbnail View Control (TVC) Zone Validation Control (ZVC)

Scanning Control

Scanner Parameter Control (SPC) Scanner Setup Control (SSC)
Visual Toolbox - Workflows
Easy to Use Workflow Assistant - One Screen Interface
Create Workflow Developer creates application to include Workflow Assistant End users can use step by step Workflow Assistant to create Workflows Execute Workflow One Workflow execution function call in application performs all processing steps
The Most Features, Formats & Destinations Unlimited Possibilities, Maximum Productivity

Workflow Benefits

Developers using OmniPage Workflow can:
Easily implement complex OCR and image processing tasks Easily manage parameters/settings in the process Easily integrate into existing applications Pass great level of flexibility to end users for customer tuning and training

End users can also benefit from OmniPage workflow:
Enjoy up to 40% performance boost Workflow execution function optimizes load balance on dual core, hyper thread, and multiple processor systems Visually track execution steps of complex process
Some of CSDK 16s New Functionalities
Some CSDK Functionalities
Preprocessing for digital camera images

Perspective distortion

Original image Output after recognizing the old way
Processed image Output after recognition
Enter the SPAMMED Architecture Framework (SAF). SAF identifies the steps and activities you need to design, model, and build successful architectures. But designing software architectures is not just the modeling. There are all kinds of activities that help ensure that the model is usable and that the architecture is successful; hence, the acronym "SPAMMED," which: Identifies Stakeholders Lists Principles, goals, and constraints Discovers quality Attributes Models Maps to technology Evaluates Deploys It is important to catalog project stakeholders, as the architect's role is to strike a balance between the (sometimes conflicting) concerns and agendas of stakeholders versus the project's functional/nonfunctional requirements. Stakeholder needs/concerns can impact the architecture. The most obvious example is a time constraint; for instance, "finish the project in two months or don't do it at all." Such constraints are bound to have an affect on the possible complexity of the architecture. The place to start is to build a mental model of the stakeholders in the project. Keep in mind that the reality is that not all stakeholders necessarily want the project to succeed. You can map the stakeholders by their interest in the project, their power, and the importance of their concerns. As a rule, you should closely manage the stakeholders who have lots of interest in the project and who can influence its success. Keep stakeholders who have low interest but high influence satisfied; keep the stakeholders who are interested but have low influence informed, and monitor the rest (see "Another View at Enterprise Architecture Viewpoints," by J. Schekkerman; www.enterprisearchitecture.info/Images/ExtendedEnterprise/E2AViewpoints_IFEAD.PDF).

Identify Stakeholders

Identifying a project's stakeholders and their concerns, agendas, and needs is the first task architects should perform when starting on new projects. Who are stakeholders? They are the people who have some vested interest in the project. Naturally there are many stakeholders with any project customers, targeted end users, operations (IT), developers, maintainers, management, testers, and so on. Stakeholders also include the architects themselves.

Warped text

Original image Processed image
Preprocessing for digital camera images How to use in CSDK16?
3D de-skew is part of preprocessing kRecSetImgDeskew(sid, DSK_3D); kRecPreprocessImg(); semi-automatic 3D de-skew in SET tools
PDF Form Data Extraction A simplified workflow for extracting PDF-based form data: RecProcessFormPagesPDF(,templateForm, pageRange, inputForms, outFile)
Fillable PDF (templateForm)
Filled forms (inputForms): PDF (filled or flat) scanned forms

CSV file (outFile)

PDF Form Data Extraction
Temporary form templates are created from each page of the sample PDF form Create text zones from text fields, list boxes and combo boxes Create OMR zones from check boxes and radio buttons Create anchor zones from static text
Processing filled PDF forms
Active PDF (filled with PDFConverter Pro or Acrobat) Simply gather the information from the PDF file Flat PDF (e.g.: printed to PDFCreate! from AdobeReader) load the PDF file apply form template find anchors (text based anchors: text search) calculate the zone positions recognize inside zones export data
Processing scanned forms (printed from AdobeReader)
load the image file apply form template recognize the whole page find anchors (text based anchors: text search) deskew the image (based on the anchors) calculate the zone positions clean BW image (dropout color, remove lines and small pieces of neighboring text) recognize inside zones export data

Short demonstration

Automatic Form Template Creation (LFR)

1. Create

Scan or load blank form; LFR creates an editable form template with comprehensive zone attributes and saves it in XML format

2. Registration

LFR automatically registers forms to be processed against template

3. Form Processing

Application calls recognition engine to recognize data fields on scanned forms

OmniPage Capture SDK v16

Thank you for your attention! Questions?

1. BACKGROUND

Optical Character Recognition (OCR) is a technology that is used to translate scanned images of text into computer editable and searchable text. Among others, the following are the major advantages of the OCR technology: It can be used to scan and preserve historical documents. It can be used for scanning data entry forms in a faster and less error prone manner. It can be used with other computer applications, such as Archives and Records Management Systems, to convert scanned documents into searchable text. At present the recognition of Latin-based characters from wellconditioned documents can be considered as a relatively feasible technology. On the other hand, the processing of non-Latin scripts is still a subject of active research. Ethiopic script-based OCR processing is currently among the least developed ICT disciplines in the country. Developments in this
area are mainly limited to preliminary research activities undertaken at different institutions of higher educations, such as the former School of Information Science for Africa (SISA). Such efforts are undertaken in an uncoordinated ways.
Development of Ethiopic Keyboard Layout and Typeface Standards EICTDA
2. OCR RESEARCHES IN Our team has tried to review the local efforts undertaken in the ETHIOPIA country in this area. At present no single production quality or

commercial documents.

application

exists

processes

Ethiopic

Some students of the departments of Information Science, Computer Science and Electrical and Computer Engineering of the Addis Ababa University have produced research output that is of interest to the fields of Optical Character Recognition Software development to Ethiopic documents. The students have used different algorithms to reach to a better result in the recognition of characters from scanned documents. Some of the algorithms and techniques that we reviewed are: Character Recognition based on polygonal
approximation (mathematical) and topological features (The Application of OCR Techniques to the Amharic Scripts: By Worku Alemu) Thinning algorithm (A generalized Approach to Optical Character Recognition (OCR) of Amharic Texts: By Million Meshesha) Training with Neural Networks ( Development of a versatile Character recognition system for Amharic texts: By Yaregal Assabie) The researches have encompassed both typewritten and computer written document types. The studies show that those tests done on the latter type of documents produce a better result. It is the recommendation of all researchers in this area that, to achieve a better result in OCR as applied to the Ethiopic script, a national standard font and keyboard mappings must
be designed and developed. This will help the test results that are obtained through the learning process (e.g. Neural Networks) to produce better results as the variation of font types and styles becomes less. Although the efforts of such researches which are undertaken as a partial fulfillment to the requirements of the various degree programmes are commendable, there is a clear lack of coordinated effort that would take such researches beyond academic exercises.

3. INTERNATIONAL MULTILINGUAL OCR TOOLS VENDORS
One of the steps that the Government may consider taking in order to develop the OCR technology is to work in partnership with international vendors that develop
multilingual OCR processing tools.
We have identified some of the major developers of OCR Software Development Kits (SDK) in order to identify the requirements of incorporating additional scripts in such SDKs. One of the major inputs demanded by such vendors is the availability of a standard Typeface that would enable the OCR tools learn the features of the Ethiopic characters. The output of this project can be considered as step forward in this regard.
The following are among the major providers of OCR Software Development Kits.

1. Microsoft

Office

Document

Imaging

Library

(MODI)
2. OmniPage Capture SDK 3. Leadtools OCR Programming Tools
The focus in this regard on the identification of Programmable tools that provide interfaces for
customization, as opposed to end-user OCR products. 3.1. Microsoft Office Document Imaging Library (MODI)
Microsoft has recently incorporated an OCR
processing technology in its MS Office 2003 packages known as Microsoft Office Document Imaging Library (MODI).

provides

Application

Development

Interfaces (API) that allows developers incorporate OCR capabilities into their products.
Although MODI does not provide support for Ethiopic, it provides support to other non-Latin scripts. The present version of MODI supports the following languages: Chinese Czech Japanese Korean Russian, and A number of European Languages
One of the advantages of MODI is that it is freely available to developers that have installed an MS Office product.

OmniPage Capture SDK

OmniPage Capture SDK is a popular product that supports OCR processing capabilities to nonEuropean languages such as the Japanese, Chinese and Korean scripts. The OmniPage Capture SDK provides different OCR engines such as print OCR (OCR, OCR-A, OCR-B and MICR), Handprint (ICR), Check Mark (OMR) and Barcode recognition engines. It also provides image file enhancement tools as well as facilities for exporting processed outputs to different formats such as PDF and XML.
Leadtools OCR Programming Tools
Leadtools OCR Programming Tools is another popular OCR SDK that supports multilingual OCR processing. It provides facilities for exporting processed output into different file formats; provides different OCR engines; and includes image file enchantment facilities.

4. COMMON STEPS OF OCR PROCESSING
The process of converting documents into electronic forms, which is usually referred to as digitization is undertaken in different steps. The process of scanning a document and representing the scanned image for further processing is called the preprocessing or imaging phase. The process of manipulating the scanned image of a document to produce a searchable text is called the OCR processing stage. 4.1. The Imaging Stage The imaging process involves scanning the document and storing it as an image. The most popular image format used for this purpose is called Tagged-Image File Format (TIFF). The resolution (number of dots per inch dpi) determines the accuracy rate of the OCR process. 4.2. The OCR Process The major steps of the OCR processing stage are shown below. 4.3. Distinguishing between text and images
Segmentation In this step, the process of identifying the text and image blocks of the scanned image is undertaken. The boundaries of each image are analyzed in order to recognize the text. 4.4. Character recognition Feature Extraction This step involves recognizing a character using a method known as feature extraction. OCR tools store rules about the characters of a given script using a method known as the learning process. A character is then identified by analyzing its shape and
comparing its features against a set of rules stored on the OCR engine that distinguishes each character. 4.5. Recognition of Words Following the character recognition process, word identification process is performed by comparing the string of characters against an existing dictionary of words. Additional processes such as spell-checking are performed under this step. 4.6. Correction of Unrecognized Characters Error
Correction In this step, the user is allowed to provide corrections to unrecognized characters. 4.7. Output Formatting The final step involves storing the output in one of the industry standard formats such as RTF, PDF, WORD and plain UNICODE text.

5. GUIDELINES FOR SELECTING AN OCR PROCESSING PRODUCT
The following are widely accepted guidelines for selecting high quality OCR processing products. The guidelines can be particularly useful if the Government considers the possibility of working with international OCR Software Development Kit vendors. The guidelines should be applied under similar conditions in terms of the quality and size of documents to be scanned, the quality of the scanning hardware and the resolution under which the document was scanned.
5.1. The Number and Type of OCR Engines Available
Modern OCR software uses multiple engines to achieve a high level of accuracy. Such engines include Handprint (ICR), machine print OCR (OCR-A, OCR-B, etc), Check Mark (OMR) and Barcode (1D and 2D) recognition engines.

5.2. Recognition Speed

The speed at which the OCR software recognizes a given scanned document is an important OCR software selection factor.
5.3. Image Testing Process
This criterion refers to the ability of the OCR software in converting printed documents to computer application files such as RTF, MS WORD, and PDF. For example, the OCR output may be saved as rtf file (Rich Text Format document) and compared to the original physical document. The number of errors in the scanned document is then calculated for each OCR software.
5.4. Supported Output Formats
The number of possible output formats such as XML, HTML, PDF, and DOC in which the output of the
scanned documents can be saved is also an important selection factor.
5.5. Support for Unicode Fonts
The OCR software under consideration should also support Unicode fonts that consist of glyphs of different international character sets including Ethiopic.
5.6. File Enhancement Features
The quality and conditions of the original documents affects the OCR processing process. The OCR software must have facilities for removing discolorations and improving contrast. Such paper color removal and contrast adjustment enhancement should not affect the accuracy level of the OCR processing.

5.7. Availability of advanced features
The availability of advanced features such as spell-checkers and WYSIWYG editors should also be considered as one of the selection criteria.
6. CHARACTER RECOGNITION USING NEURAL NETWORKS
Although, there exist a number of methodologies of addressing the issues of character recognition, applying Artificial Intelligence Neural Networks is nowadays the most popular method for recognizing non-Latin scripts such as Ethiopic. This is a research area that the EICTDA may consider working with researchers from the various academic institutions as well as other private bodies. Neural Network is one of the areas of Artificial Intelligence that has gained popularity in multilingual Optical Character Recognition. In particular, the Backpropagation neural network is one of the popular approaches employed for this problem area. A typical neural network for such implementations consists of three interconnected layers Input, Hidden and Output layers. The processing nodes of each layer are tied together with weighted links. The main approach is that a training set is first prepared, and then a neural network is trained to recognize patterns from the training set. The training process teaches the network to respond with desired output for a specified input. Each training sample is represented by possible input and the desired network's output for the input. Finally, the network is expected to produce the desired output for a given arbitrary input. To apply the process to character detection, the training process involves storing a large number of character images with their corresponding Unicode character. The training set is used during the character recognition stage of the OCR process. The above discussion of the Neural Networks approach provides only a brief outline of the overall methodology. We recommend that the EICTDA support through researches in this area. We have annexed a research paper in this regard to demonstrate that there exist promising efforts at the research level, which could be promoted to production level outputs.

7. RECOMMENDATIONS

Even though the OCR technology is one of the important areas of the Computer technology, our assessment reveals that very minimal effort is being undertaken in this area. Thus the Ethiopian Government
should put efforts towards the development and promotion of the OCR technology.
Apart from the very few graduation thesis outputs, there exists on production quality Ethiopic OCR application. To promote the development of the Ethiopic OCR technology, the following recommendations are provided. 7.1. Enforce the use of national standard typeface. The use of a standard typeface across organizations would make the processing of Ethiopic documents much easier, as the attributes of such typeface would be pre-known to OCR processing applications. Thus Government has to undertake promotional activities regarding the designed standard typeface and eventually enforce the use of the standard typeface for the exchange of documents among Governmental organizations. 7.2. Support OCR Research Activities. In order to achieve a sustainable outcome, the Government has to consider this area as one of its priority research components. A coordinating body should be setup to oversee the research outcomes undertaken by academic institutions, private bodies as well as Governmental intuitions. A public resource center should also

be setup to promote and disseminate the outcomes of the research activities. 7.3. Outsource International OCR Software
Development Kits. At present there exist a number of international OCR Software Development Kit (SDK) vendors that support the processing of Multilingual Documents, including documents created using Japanese, Chinese and Korean scripts. The EICTDA needs to work in partnership with the International vendors in order to incorporate the Ethiopic script in such products. The availability of a standard typeface and keyboard layout and the fact that the Ethiopic script is part of the Unicode standard character set is expected to make the inclusion of Ethiopic OCR processing in the popular products far easier. The selection criteria presented in section 5 should be used to select the most appropriate product.
Annex 1: Ethiopic Ocr Research Paper
Unicode Optical Character Recognition

By Daniel Admassu

http://www.codeproject.com/cs/algorithms/UnicodeOCR.asp

Contents

1. 2. 3.
Background... 1 OCR Researches in Ethiopia.. 2 International Multilingual OCR Tools Vendors. 4 3.1. 3.2. 3.3. Microsoft Office Document Imaging Library (MODI). 4 OmniPage Capture SDK... 5 Leadtools OCR Programming Tools.. 6
Common Steps of OCR Processing.. 7 Guidelines for Selecting an OCR Processing Product. 9 5.1. 5.2. 5.3. 5.4. 5.5. 5.6. 5.7. The Number and Type of OCR Engines Available. 9 Recognition Speed.. 9 Image Testing Process.. 9 Supported Output Formats. 9 Support for Unicode Fonts.. 10 File Enhancement Features.. 10 Availability of advanced features. 10
Character Recognition using Neural Networks. 11 Recommendations.. 12
Annex 1: Ethiopic OCR Research Paper.. 14

Part III

Ethiopic OCR
(Optical Character Recognition)