Pre-built Recognizers

Overview

The Lipi Alphanumeric Character Recognizer Package lipi-reco-char-alphanum contains separate “pre-built” recognizers for isolated handwritten English uppercase characters, lowercase characters and numerals in digital ink form captured using an electronic stylus. They are built using the Lipi Core Toolkit and are meant for users interested in integrating such recognition capabilities into an application.
These recognizers are not meant to represent the state of the art in recognition technology, but rather provided for the purpose of creating interesting new applications.

The recognizer package comes with three pre-built shape recognizers:
► numerals : For Indo-Arabic numerals [0-9].
► eng_upper : For uppercase English alphabet [A-Z].
► eng_lower : For lowercase English alphabet [a-z].

Package Contents

The package contains dynamic libraries, configuration files and model data files for the three recognizers.

Supported Platforms

lipi-reco-char-alphanum supports the following 32 bit platforms:
► Windows XP Professional Edition
► Red hat Enterprise Linux Edition 4.0
► Ubuntu Hardy 8.04

System Requirements

Disk Requirements
lipi-reco-char-alphanum has packages for Windows and Linux. In the case of Windows, separate packages are provided for VC6.0 and VC2005.

Package	Package Size	Disk space required
lipi-reco-char-alphanum (Windows XP Professional, VC6.0 )	8.18 MB	21.10 MB
lipi-reco-char-alphanum (Windows XP Professional, VC2005 )	12.20 MB	38.20 MB
lipi-reco-char-alphanum (Linux)	9.99 MB	27.40 MB

Software Requirements
Windows
► Microsoft Visual Studio 6.0 with SP6 or Microsoft Visual C++ 2005 ( http://msdn.microsoft.com/vstudio/)
► Cabarc sdk (required for packaging only) ( http://support.microsoft.com/default.aspx?scid=kb;en-us;310618)
► Perl 5.0 or above ( http://www.activestate.com/Products/ActivePerl/ )

Linux
► GCC 3.3.3 ( http://gcc.gnu.org/)
► tar
► Perl 5.0 or above

Recognition Accuracy

The recognizers use Nearest Neighbor shape recognition method and Dynamic Time Warping (DTW) distance as the measure of similarity. Recognizers have been trained and benchmarked using the IRONOFF handwriting dataset. For more details please see the lipi-reco-char-alphanum User Manual.

Recognizer	Accuracy Reported
numerals	97.58%
eng_upper	93.72%
eng_lower	90.03%

Compatibility with Lipi Core Toolkit

The recognizers are built using the version 2.2 of Lipi Core Toolkit.

User Manual

A detailed description of the usage of the lipi-reco-char-alphanum is explained in the User manual page.

Latest Release

lipi-reco-char-alphanum 2.1 / 2.2 is the latest released version of the core toolkit. For more details, please visit the lipi-reco-char-alphanum release page.

Acknowledgements

We would like to thank IRESTE, University of Nantes (France) for allowing the use of the IRONOFF handwriting database for training the pre-built recognizers.