A computationally efficient technique for discriminating between hand-written and printed text

During the processing of documents from a range of sources, it is useful to be able to discriminate documents with hand-written text from those with printed text. This allows the two to be processed in different ways; for example the typed text using a high speed automated OCR reader and the hand-written text manually, so optimising the use of the expensive OCR reading machine. The paper describes a computationally efficient technique for discriminating between hand-written and printed text on mail. The method employs two stages of processing. The first involves the extraction of a number of low-level features from an image of the sample text, while the second stage is a parametric classification operation, which employs a relatively simple feedforward multilayer perception neural network. The processing involved in each of these stages is described in detail. In addition, results which were obtained when using the optimised techniques to discriminate between hand-written and printed addresses are presented.