Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BeHGAN: Bengali Handwritten Word Generation from Plain Text Using Generative Adversarial Networks

Md. Rakibul Islam, Md. Kamrozzaman Bhuiyan, Safwan Muntasir, Arifur Rahman Jawad, Most. Sharmin Sultana Samu

TL;DR

BeHGAN addresses Bengali word-level handwritten text generation from plain text by introducing a semi-supervised GAN that generates per-character handwriting with a shared style vector. A new Bengali handwriting dataset from ~500 participants and a dataset-augmentation pipeline support training and evaluation. The model demonstrates moderate fidelity (SSIM 0.67, FID 41, Geometric Score 0.63) and highlights the importance of data diversity and post-processing (e.g., GFP-GAN) for HTG. Limitations include restricted alphabet coverage and low output resolution, with future work targeting broader Bengali alphabets, conjunct characters, and diffusion-based improvements.

Abstract

Handwritten Text Recognition (HTR) is a well-established research area. In contrast, Handwritten Text Generation (HTG) is an emerging field with significant potential. This task is challenging due to the variation in individual handwriting styles. A large and diverse dataset is required to generate realistic handwritten text. However, such datasets are difficult to collect and are not readily available. Bengali is the fifth most spoken language in the world. While several studies exist for languages such as English and Arabic, Bengali handwritten text generation has received little attention. To address this gap, we propose a method for generating Bengali handwritten words. We developed and used a self-collected dataset of Bengali handwriting samples. The dataset includes contributions from approximately five hundred individuals across different ages and genders. All images were pre-processed to ensure consistency and quality. Our approach demonstrates the ability to produce diverse handwritten outputs from input plain text. We believe this work contributes to the advancement of Bengali handwriting generation and can support further research in this area.

BeHGAN: Bengali Handwritten Word Generation from Plain Text Using Generative Adversarial Networks

TL;DR

BeHGAN addresses Bengali word-level handwritten text generation from plain text by introducing a semi-supervised GAN that generates per-character handwriting with a shared style vector. A new Bengali handwriting dataset from ~500 participants and a dataset-augmentation pipeline support training and evaluation. The model demonstrates moderate fidelity (SSIM 0.67, FID 41, Geometric Score 0.63) and highlights the importance of data diversity and post-processing (e.g., GFP-GAN) for HTG. Limitations include restricted alphabet coverage and low output resolution, with future work targeting broader Bengali alphabets, conjunct characters, and diffusion-based improvements.

Abstract

Handwritten Text Recognition (HTR) is a well-established research area. In contrast, Handwritten Text Generation (HTG) is an emerging field with significant potential. This task is challenging due to the variation in individual handwriting styles. A large and diverse dataset is required to generate realistic handwritten text. However, such datasets are difficult to collect and are not readily available. Bengali is the fifth most spoken language in the world. While several studies exist for languages such as English and Arabic, Bengali handwritten text generation has received little attention. To address this gap, we propose a method for generating Bengali handwritten words. We developed and used a self-collected dataset of Bengali handwriting samples. The dataset includes contributions from approximately five hundred individuals across different ages and genders. All images were pre-processed to ensure consistency and quality. Our approach demonstrates the ability to produce diverse handwritten outputs from input plain text. We believe this work contributes to the advancement of Bengali handwriting generation and can support further research in this area.
Paper Structure (14 sections, 7 figures, 3 tables)

This paper contains 14 sections, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Background Study
  4. Semi-Supervised Generative Adversarial Network (SGAN)
  5. Generative Facial Prior-GAN
  6. Evaluation Metrics
  7. Methodology
  8. Model Architecture
  9. Generation Workflow
  10. Experimental Setup
  11. Dataset Collection and Preprocessing
  12. Dataset Organization
  13. Result Analysis
  14. Conclusion and Future Works

Figures (7)

  • Figure 1: Word Generation Architecture (inspired by fogel2020scrabblegan). The input Bengali word is mapped to the English letters "klm". The model selects the corresponding character samples from the dataset and generates a mask for each character. Due to the overlapping receptive field property of CNNs, the generator produces each character individually. The generated characters are then combined with overlapping regions to form the final handwritten word.
  • Figure 2: Proposed Methodology
  • Figure 3: Data Collection Format
  • Figure 4: Before and After applying filter
  • Figure 5: Individual word samples after cropping
  • ...and 2 more figures