The A 2 iA Historical Document Local Attribute Detection System submitted to the ICFHR-2016 ANDAR-AD competition

This paper describes the local attribute detection systems proposed by A2iA to the ICFHR-2016 ANDAR-AD competition. These systems are based on Recurrent Neural Networks (RNN), originally trained as classifiers to recognize 7 different attributes of historical documents images. Both the provided development data and additional proprietary data were used to train the classifiers.