License Plate Recognition Using Weighted Finite-State Transducers and Different Evidence Combination Methods

In this paper, a method to post-process the results of a License Plate Recognition (LPR) is proposed, based on Weighted Finite-State Transducers (WFSTs), along with the extension of the technique to a sequence of images of the same vehicle. We use the LPR output with a posteriori class probabilities, an error model with symbol substitutions, insertions and deletions, and a constraint model encoding the valid formats of the plates, to build three different WFSTs which are then composed. The results of this method is compared with other approaches for a dataset taken in a real parking lot facility.