Effect of Neural Network Generalization on the Online Handwritten Patterns

Abstract

The work focusses on the recent trends on pen computing applications. Online handwritten patterns are acquired using pen tablets. In Online handwritten systems, data is acquired during the writing, which provides the dynamic movements of the pen trajectory with the time. The acquired online patterns are preprocessed, and angular information for the successive points are extracted. These features are input to the Feed Forward Neural Network model. Supervised Learning methods are implemented, and the performance of the system is evaluated for the generalization capability of the Neural Network on the online handwritten patterns. The selected patterns are vertical, horizontal and with various angular strokes. The evaluation results demonstrate the robustness of Neural Network for online pen strokes. The Neural network exhibits significant generalization for unseen, a new set of data.