Automating performance optimization for script word recognition systems

Abstract

In this paper we present a system for automating parameter optimization of pattern recognition systems and demonstrate its capabilities on script word recognition systems. This system, called Optima (optimization manager) has been specially developed to fit the requirements of optimization in the pattern recognition field, where computation- and engineering-costs for system evaluation are very high. Our experiments show that automatic parameter optimization not only performs the task as well as the experienced engineers, thereby relieving them from routine work, but ultimately also outperforms hand-tuning in terms of system performance.