Mobile banknote recognition: topological models in scene understanding

Portable assistive devices are getting more and more widespread. Since most cell phones are nowadays equipped with cameras they can provide an easy way for blind and visually impaired people to obtain visual information about their environment. In this paper we present a banknote recognition system for our previously proposed mobile device, called the bionic eyeglass. We employ a recognition scheme based on the structural topology of banknotes that enables that the visual features detected are not joined in an ensemble classifier independently, but they are coupled via an excitation-inhibition model. We have shown that the technique outperforms classical ensemble classifiers.