Optical Character Recognition (OCR)

Key Attributes and Limitations

	Data is in human readable form
	Progressive area of optical-read technology, characterised by systems capable of high speed, accurate recognition, capability of handling multiple fonts and distorted characters, but with level of performance naturally reflected in cost
	Low cost, PC-based systems available for document management applications
	Variety of character formation techniques, primarily printer-based technology, with costs determined by the type and quality of printer.
	Generally close proximity scanning required to capture images
	No encoded error control (error detection and correction) - reliant upon processing capability for recognising characters

Forty years ago, before bar code technology was a gleam in the grocery industry's eye, OCR was being used in commercial applications. The technology was initially designed to read highly stylized human-readable fonts, such as OCR-A, which encodes the alphanumeric character set as well as 60 other shapes.

In 1975, OCR was adopted by the National Retail Merchants Association (now known as the National Retail Federation, or NRF) as the standard font for merchandise identification, credit authorization, and inventory control. However, poor supplier source marking as well as unreliable scanning equipment prompted a shift in the 1980s to barcode source marking of general merchandise, which proved to be much more successful.

The evolution of high-powered desktop computing has benefited OCR reading technology over the last few years, allowing for the development of more powerful recognition software that can read a variety of common printer fonts. High-end systems use sophisticated neural networks, which enable the system to improve its read accuracy over time by learning the nuances of a particular font and even varying styles of unconstrained handwriting. Most OCR systems today are font-independent and are available in three different configurations: page readers, transaction readers (usually numerical only), and handheld readers.

Characters are scanned with a light source, providing an image that is interpreted by the recognition software. The software uses one of two approaches to character analysis: template matching (whereby the character is matched to a database of possibilities) and feature extraction (analyzing structural elements of the character).

OCR shines in applications where human-readability is required, in electronic document processing and management, and in high-volume scanning of numerical transaction data. Neural net-based OCR systems are making headway in reading unconstrained handwriting, but such systems are as yet substantially inaccurate and prohibitively expensive.

For large commercial applications, font-independent OCR systems are considerably less accurate than those dedicated to an OCR font, and even a dedicated OCR system is less accurate than a barcode-based system. For this reason OCR technology is not likely to have a great impact on AIDC applications. However, OCR will continue to serve and grow in its established niches, particularly in electronic document processing and management applications, and in those industrial applications where the barcode's lack of human-readability rules it out.