Abstract
In the past few years, there have been massive developments in the field of Deep Learning and Neural Networks, this work intends to leverage these recent developments for the use of Handwriting Generation. Input for handwriting data can be taken in two forms, off-line and on-line. Off-line format of handwriting data consists of input in the form of images usually in gray-scale. Whereas on-line format consists of time series data having coordinates and a timestamp associated with multiple input points which together constitute a character and ultimately assimilate to form a word. Generally Recurrent Neural Network (RNN), Long Short Term Memory (LSTM) and Mixture density networks (MDN) as used to train a model to probabilistically predict the direction of the next stroke from the string input and the previous data point coordinates. This method, though effective, has a greatly reduced scalability as the input collected is restricted to only the handwriting collected via digital writing boards. There have been many advances in the field of Deep Learning and Generative Neural Networks. Our proposed work utilizes Deep Convolutional Neural Networks in the form of conditional Deep Convolutional Generative Adversarial Networks, to overcome the shortcomings, and expand the scalability of our model to accept documents and pictures as handwriting input, greatly increasing the scalability of our system.