Abstract
Pattern recognition using statistical models such as Dynamic Bayesian Networks (DBNs) is currently a growing area of study. The classification performances typically greatly rely on the adequation between the data and a DBN model, the latter having to best describe the dependencies in each class of data. In this paper, we present a new approach based on optimising the sequences and layout of observations of DBN models in a hierarchical Bayesian framework, applied to the classification of handwritten digit. Classification results are presented for the described models, and compared with previously published results from probabilistic models. The new approach was found to improve the recognition rate compared to previous results, and is more suitable for applications were a high speed in the recognition phase is important.