In recent the new requirements for automation of bank payment documents are emerging. In processing of bank checks which are the most widespread payment instruments it is required not only to recognize payment amount but also to recognize the names of the payer and the recipient, verify signatures, addresses, bank accounts, code strings. The three different processes of automated checks reading: recognition, detection and verification. The one of the most complex tasks in check processing is in localization of required information fields on the image of document due to typical weak structuring of bank checks. The processing begins with segmentation stage which is being followed by symbol and word recognition stage which finally ends in filtration by dictionary.