PC-based machine vision systems are becoming more powerful as the individual components that comprise them increase in performance. This has allowed for more complex algorithms to be developed that can learn, see, identify, and guide. Learning algorithms for machine vision are essential for many of the applications that require inspection and guidance. These algorithms typically require some type of learning step before the actual inspection can take place. Some of the algorithms that fall into this category include Optical Character Recognition (OCR), pattern matching, and color matching, as previously reported in "PC-Based Software for Pattern, Color, and Color Pattern Matching", Photonics Tech Briefs, NASA Tech Briefs Vol. 27, No. 6 (June 2003), page 10a. Many industries rely on these algorithms to speed-up manufacturing, increase yields, and improve quality.

OCR applications include inspecting pill bottle labels, verifying wafer lot codes, identifying stamped machine parts, and sorting of mail packages and parcels. Many of the characters are applied in an easily readable fashion such as ink jet or label printing, but a large portion is machine stamped or laser etched onto parts. Etched characters can be applied on materials such as paper, aluminum, plastic, or glass. The combination of application methods and materials where the text is applied presents a challenge to a computer to correctly identify the characters.

OCR applications aid users by not limiting them to a single, predefined font. This software recognizes unique fonts and characters on labels such as those used on pharmaceutical and consumer products.

This PC-based trainable optical character recognition software, called NI OCR Software, is specifically designed for machine vision application developers under pressure to reduce costs and deliver products quickly. The software provides high-speed, reliable reading performance, despite poor and inconsistent image quality that may result from industrial process variations.

NI OCR Software can be taught to process characters from a wide variety of fonts, symbols, and characters. By using the training engine and intuitive training interface, users are not limited by any predefined fonts, although the software is suitable for industrial fonts such as OCR A, OCR B, and SEMI.

The software also includes a large collection of parameters with which individual characters can be segmented and learned. Users can alter the built-in segmentation routines to distinguish between individual characters even under the most difficult imaging conditions. Segmentation parameters include minimum and maximum height and width and horizontal and vertical spacing options. Also, automatic thresholding ensures that characters are identified properly. In addition, users can detect characters that may change in scale.

This work was done by Jason Mulliner, Vision product manager, for National Instruments. For more information call (512) 683-0100 or visit National Instruments online at www.ni.com. NI OCR Software is an add-on package to the NI Vision Development Module.