Brazilian bank check handwritten legal amount recognition

This paper presents a system that is being developed for the recognition of the handwritten legal amount in Brazilian bank checks. Our strategy used to approach the handwritten legal amount recognition problem puts on evidence the keywords: "mil", "reals/real", "centavos/centavo" which are almost always present in each amount. The recognizer, based on hidden markov models, does a global word analysis, therefore, it does not carry out an explicit segmentation of words into characters or pseudo-characters. In this context, each word image is transformed into a sequence of observations using pre-processing and feature extraction stages. Our system, when tested on our database simulating Brazilian bank checks, shows the viability of our approach.