Off-line cursive handwriting recognition using synthetic training data

The objective of this thesis is to investigate the generation and use of synthetic training data for off-line cursive handwriting recognition. It has been shown in many works before that the size and quality of the training data has a great impact on the performance of handwriting recognition systems. A general observation is that the more texts are used for training, the better recognition performance can be achieved. In this work it is examined whether this observation holds if the training set is augmented by synthetically generated texts. The motivation is that augmenting the training set by computer generated text samples is much faster and cheaper than collecting additional human written samples. For this purpose, two novel methods are presented for the generation of synthetic text lines. The first one is based on the geometrical perturbation of existing human written text line images. In the second method, handwritten text lines are synthesized from ASCII transcriptions, using templates of characters and the Delta LogNormal model of handwriting generation. To evaluate these two methods for synthetic training set expansion, the task of off-line cursive English handwritten text line recognition is considered. In the last part of the thesis, a novel approach for text line segmentation into individual words is presented, in order to perform segmentation-based text line recognition, which gives new insights into the effects of synthetically expanded training sets. Several configurations of the recognizer and the synthetic handwriting generation process are examined in the thesis. Based on the experimental results, it is concluded that the use of synthetic training data can lead to improved recognition performance of handwriting recognition systems.