Abstract

In this thesis, an automatic system for processing date information handwritten on bank cheques is described. This system is composed of date image segmentation, handwritten digit recognition, and cursive word recognition. The proposed method does not impose any restriction or require any a priori information on the date written, and is able to handle both English and French cheques. With appropriate modification, this method can be extended to recognize date in other application environments. Much effort has been dedicated to the design and implementation of date image segmentation. Two groups of features, shape features and spatial features, are developed and utilized to locate possible cutting positions on date$\sb{-}$zone images. In addition, a set of heuristic rules are proposed to complete the segmentation of this image, identify the three fields (Day, Month and Year), and also determine the writing style of Month in order to apply the appropriate recognizer (i.e. digit or cursive word) to process the Month field. Both segmentation and date recognition systems were tested on two different databases, i.e. (a) the CENPARMI database, and (b) data collected from a utility company. The results obtained prove the efficiency and feasibility of the proposed date recognition system. Areas which require further development are also specified. An image tagging tool has been developed for this research. In addition to its user-friendly environment and simple operation, it also provides powerful functionality in recording contextual information. This makes it possible and convenient to extract data such as words, digits, characters, punctuations, etc. A lot of useful statistics can also be gathered from the tagged information