NIST Form-Based Handprint Recognition System

The National Institute of Standards and Technology (NIST) has developed a new release of a standard reference form-based handprint recognition system for evaluating optical character recognition. As with the first release, NIST is making the new recognition system freely available to the general public on CD-ROM. This source code testbed, written entirely in C, contains both the original and the new recognition systems. New utilities are provided for conducting generalized form registration, intelligent form removal with character stroke preservation, robust text-line isolation in handprinted paragraphs, adaptive character segmentation based on writing style, and sophisticated MultiLayer Perceptron (MLP) neural network classification. A software implementation of the machine learning algorithm used to train the new MLP is included in the test-bed, enabling recipients to train the neural network for pattern recognition applications other than character classification. A host of data structures and low-level utilities are also provided. These include the application of spatial histograms, affine image transformations, simple image morphology, skew correction, connected components, Karhunen Loève feature extraction, dictionary matching, and many more. The software test-bed has been successfully compiled and tested on a host of UNIX workstations including computers manufactured by Digital Equipment Corporation, Hewlett Packard, IBM, Silicon Graphics Incorporated, and Sun Microsystems.1 Approximately 25 person-years have been invested in this software test-bed, and it can be obtained free of charge on CD-ROM by sending a letter of request via postal mail or FAX to NIST. This report documents the new recognition software test-bed in terms of its installation, organization, and functionality.