Research On Check Recognition

Information technology facilitates the new business growth of financial industry in terms of the application of electronic products. How to process the daily business of the banks by using information technology quickly and accurately, shorten the time of business operation, keep customer primitive information, and carry on electronic authentication, becomes the research subject of bank automation.Automatic processing of bank checks is usually composed of four steps. Firstly, check images are scanned into computer by input devices such as scanner and vidicon etc. and stored as image data. Next work is to fetch check images from storage devices and recognize the types of the images. Thirdly, localization and cutting of the interested information such as printed and written characters in check images is to be done. Since the regions cover interested information have fixed positions in each check image, it is easy to localize and cut the interested information after check type recognition. Finally, interested information such as user name, accounts, sum, operator, date in the cut regions is recognized and stored into database.Among all the processing steps, financial documents recognition is in the first place. This paper proposes an algorithm of financial document recognition based on characteristic line detection.In the field of image pattern recognition, it is of great importance to choose what kind of characters of the image as well as how to extract character from an image. The recognition can not be accurate unless the character of the image is extracted properly. Most of the check images are table-structured. Usually the tables are composed of some horizontal and vertical lines-we call them characteristic lines. In different check images the number, dimension and position of the characteristic lines are always different, according to which, we set these lines as check image character for recognition.The algorithm consists of two major parts. Firstly, a line detection algorithm applied in financial document images is proposed according to the specific characters of financial documents' structure. Secondly, a financial document pattern recognition algorithm based on the fact that different financial documents have different geometrical structures is proposed.The experimental result demonstrates the reliability of the algorithms and the high speed of the operation. It also shows that the algorithms have enough identification to be put into practice.