Combining on-line and off-line bidirectional long short-term memory networks for handwritten text line recognition
Marcus Liwicki, Horst Bunke · Bern Open Repository and Information System (University of Bern) · 2008
In this paper we present a multiple classifier system (MCS) for on-line handwriting recognition. The MCS combines several individual recognition systems based on bidirectional long short-term memory networks. To obtain diverse recognizers, we use different feature sets based on on-line and off-line features. Furthermore, we generate a number of different recognizers by changing the initial-ization of the networks. To combine the word sequences output by the recognizers, we incrementally align these sequences using the recognizer output voting error reduc-tion framework (ROVER). For deriving the final decision, different voting strategies are applied. The best combina-tion ensemble has a recognition rate of 83.64 %, which is significantly higher than the 81.26 % achieved by the best individual classifier. 1.