Improved Modeling in Handwriting Recognition

In this work a script independent handwriting recognition system is proposed which is derived from the RWTH-ASR hidden Markov model (HMM) based speech recognizer. Most problems occurring in handwriting recognition (HWR) are induced by large variations within the written text. In particular, different handwriting styles such as cursive writing or long drawn-out strokes are difficult to model. Common handwriting recognition systems use various preprocessing and feature extraction methods to compensate for the variations in handwriting. Additional approaches have been made using writer adaptive training for writer dependent modeling of the variations. This work uses another approach by dealing with these variations using explicit background modeling, improving the visual models of the handwriting and exploiting the character context within the script. Therefore, only simple appearance based features and only few preprocessing steps have to be used. Instead, methods known from handwriting recognition, such as model length estimation and discriminative training for writer independent modeling as well as common methods from speech recognition are applied. In addition, several approaches for improvement of continuous text line recognition are made by using lexica and language models accounting for the specialties in line recognition. The script independence of the proposed system is demonstrated on Arabic and Latin recognition tasks. The performance is evaluated on the IFN/ENIT database which provides an Arabic single word recognition task. In addition, the IAM database is chosen, providing a Latin text line recognition task. The results obtained on both databases outperform the currently known best error rates achieved with a single recognition system.