Data Incubation -- Synthesizing Missing Data for Handwriting Recognition

TL;DR

This paper introduces a controllable handwriting synthesizer to generate diverse training data, significantly improving online handwriting recognition accuracy by combining real and synthetic samples.

Contribution

It presents a novel framework for synthesizing handwriting data with controllable content and style, enhancing recognition performance beyond real data alone.

Findings

66% reduction in Character Error Rate

Improved recognition of underrepresented content and styles

Effective data synthesis framework for handwriting recognition

Abstract

In this paper, we demonstrate how a generative model can be used to build a better recognizer through the control of content and style. We are building an online handwriting recognizer from a modest amount of training samples. By training our controllable handwriting synthesizer on the same data, we can synthesize handwriting with previously underrepresented content (e.g., URLs and email addresses) and style (e.g., cursive and slanted). Moreover, we propose a framework to analyze a recognizer that is trained with a mixture of real and synthetic training data. We use the framework to optimize data synthesis and demonstrate significant improvement on handwriting recognition over a model trained on real data only. Overall, we achieve a 66% reduction in Character Error Rate.