An ebedded OCR software architecture for enhancing portability

Abstract

A software architecture style for OCR to increase portability is presented. A hybrid style of four layers, combined with a pipe & filter and layer-type architecture is adopted to reduce the platform dependency keeping the cascading process stream of general OCR. The OCR process constructs a top layer to accomplish the main task. Data manager is separated as a different layer from the top layer, which handles input and knowledge data of the classifier. To provide a unified API, an interface layer is included. A neural network based general OCR source code is totally re-written on the foundation of the proposed style. As a result, confirmed a high portability rate over 99% of source code reuse without modification and performance sacrifices is achieved in the experiment.