A non contextual approach for textual element identification on bank cheque images

In many research papers concerning the identification of handwriting in bank cheque images, one can note the current use of contextual information. However, it is very functional to observe the behaviour of textual elements in order to distinguish between them. We present an approach based on the characterization of textual elements, whether they be handwritten or machine printed, without being becoming dependant on using elements strongly attached to their environment. Textual elements are observed in small sections and a set of features is extracted serving as the input to a classifier able to identify, the class of each input vector. Images from different banks were tested and the results prove the ability of this methodology to operate over a wide variety of documents.