Generating Synthetic Data for Text Recognition

TL;DR

This paper presents a method for generating a large synthetic handwritten dataset using open source fonts and augmentation, aiming to improve handwritten word recognition and spotting.

Contribution

It introduces a new large-scale synthetic handwritten dataset and a framework for generating realistic handwritten images for training deep learning models.

Findings

Released 9 million synthetic handwritten word images.

Enhanced training data for improved recognition performance.

Facilitated advancements in handwritten word spotting.

Abstract

Generating synthetic images is an art which emulates the natural process of image generation in a closest possible manner. In this work, we exploit such a framework for data generation in handwritten domain. We render synthetic data using open source fonts and incorporate data augmentation schemes. As part of this work, we release 9M synthetic handwritten word image corpus which could be useful for training deep network architectures and advancing the performance in handwritten word spotting and recognition tasks.