Evolutionary Tuning of Compound Image Analysis Systems for Effective License Plate Recognition

This paper describes an evolutionary algorithm applied to tuning of parameters of license plate detection systems. We consider both simple and compound detection systems, where the latter ones consist of multiple simple systems fused by some aggregation operation (weighted sum or ordered weighted average). With the structure of a system given by a human and fixed, we perform an evolutionary search in the space of possible parameter combinations. Several simple and compound structures are considered and verified experimentally on frame collections taken from highly heterogeneous video sequences acquired in varying conditions. The obtained results demonstrate that all considered systems can be effectively tuned using evolutionary algorithm, and that compound systems can outperform the simple ones.