SBIR-STTR Award

We propose to demonstrate the feasibility of Optical Character Recognition (OCR) on a vlsi neurocomputer. The OCR system will use state of the art artificial neural network classifiers on adaptive solutions' cnaps neurocomputer chips. The cnaps chips offer unprecedented performance of artificial neural networks. The image preprocessing of the ocr system will also execute on the ACNAPS chips. The goal of the Phase I research is to implement the OCR software on acnaps development system, analyze I/O and system performance, and create a board level architecture for an OCR system. The goal of the Phase II research is to design and build the OCR system hardware. Anticipated benefits/potential commercial applications - optical character recognition technology is useful to the department of defense for the computer aided logistics program, electronic backup of existing documents, and forms processing.

The objectives of the Phase II project are to build a hardware prototype of an Optical Character Recognition (OCR) system based on Adaptive Solutions' CNAPS architecture, and to implement OCR recognition software to run on the hardware. The CNAPS architecture is designed for highly parallel implementation of Artificial Neural Networks (ANNs), but is also useful as a parallel digital signal processor. The OCR software will recognize hand printed characters from forms. The recognition software will extract a field from a document, isolate individual characters, and classify them. An ANN will be used for classification of the characters. The majority of the image preprocessing, ANN classification, and context post processing will run on the CNAPS system. Anticipated

Benefits:
Optical Character Recognition is a key technology for office automation and document processing systems. Applications for hand written forms processing include requisition forms, inventory record keeping, bank checks, sales slips, and insurance forms. Key Words: Neural Networks, OCR, Parallel

Keywords:
Neural Networks, Ocr, Parallel