Character recognition methods including locating and extracting predetermined and apparatus data from a document

Abstract From the Canadian patent

A method for automatically locating and recognizing amount data (10b) on a document (10) by scanning the document to first locate a particular symbol such as a, dollar sign "$" (10a), and then using the located "$" to determine the document area (10b-1, 10b-2) containing the amount data (10b) to be recognized. Elements of the "$" symbol and amount characters are extracted from a gray-level image (Fig. 5) using binary seed and mask images (Figs 7, 8, and 10, 11) of different thresholds in a manner such that black pixels found in the higher threshold seed images are propagated in the lower threshold mask image to form connected components (Fig. 8, cc-1 and cc-2).