Automatic Processing of Brazilian Bank Checks
Rafael Palacios, Anshu Sinha, Amar Deep Gupta · 2002
This paper focuses on determining the value of the check, which is the first step needed for automatic processing. The multi–staged approach proposed in this paper deals with four aspects: the detections of the numerical amount field on the check; the segmentation of the string into a set of characters; the recognition of isolated characters; and the syntactic verification of each output. All modules have been tunned to recognize Brazilian checks. Through the use of feedback mechanism between recognition and segmentation, the adoption of a new normalization approach, the use of several neural networks in parallel, and the incorporation of syntactic verification methods, our research team has been able to significantly enhance the overall accuracy levels.