Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HCR-Net: A deep learning based script independent handwritten character recognition network

Vinod Kumar Chauhan, Sukhdeep Singh, Anuj Sharma

TL;DR

HCR-Net addresses the need for a script-independent approach to handwritten character recognition by combining partial transfer learning from a pre-trained VGG16 with image augmentation in a CNN architecture that operates on 32×32 grayscale offline images. It trains in two phases, freezing lower, pre-trained layers in Phase 1 and unfreezing all layers in Phase 2, which yields fast convergence and robust generalization across 41 datasets spanning multiple scripts. The model establishes numerous new benchmarks (up to 26) and delivers improvements up to 11% over prior results while reducing trainable parameters by about 34% relative to the full VGG16 baseline, with public code released for reproducibility. Overall, HCR-Net demonstrates strong cross-script performance and practical potential for script-independent handwriting recognition, while noting areas for future expansion to large-character-set languages and hierarchical architectures.

Abstract

Handwritten character recognition (HCR) remains a challenging pattern recognition problem despite decades of research, and lacks research on script independent recognition techniques. {\color{black}This is mainly because of similar character structures, different handwriting styles, diverse scripts, handcrafted feature extraction techniques, unavailability of data and code, and the development of script-specific deep learning techniques. To address these limitations, we have proposed a script independent deep learning network for HCR research, called HCR-Net, that sets a new research direction for the field. HCR-Net is based on a novel transfer learning approach for HCR, which \textit{partly utilizes} feature extraction layers of a pre-trained network.} Due to transfer learning and image augmentation, HCR-Net provides faster and computationally efficient training, better performance and generalizations, and can work with small datasets. HCR-Net is extensively evaluated on 40 publicly available datasets of Bangla, Punjabi, Hindi, English, Swedish, Urdu, Farsi, Tibetan, Kannada, Malayalam, Telugu, Marathi, Nepali and Arabic languages, and established 26 new benchmark results while performed close to the best results in the rest cases. HCR-Net showed performance improvements up to 11\% against the existing results and achieved a fast convergence rate showing up to 99\% of final performance in the very first epoch. HCR-Net significantly outperformed the state-of-the-art transfer learning techniques and also reduced the number of trainable parameters by 34\% as compared with the corresponding pre-trained network. To facilitate reproducibility and further advancements of HCR research, the complete code is publicly released at \url{https://github.com/jmdvinodjmd/HCR-Net}.

HCR-Net: A deep learning based script independent handwritten character recognition network

TL;DR

HCR-Net addresses the need for a script-independent approach to handwritten character recognition by combining partial transfer learning from a pre-trained VGG16 with image augmentation in a CNN architecture that operates on 32×32 grayscale offline images. It trains in two phases, freezing lower, pre-trained layers in Phase 1 and unfreezing all layers in Phase 2, which yields fast convergence and robust generalization across 41 datasets spanning multiple scripts. The model establishes numerous new benchmarks (up to 26) and delivers improvements up to 11% over prior results while reducing trainable parameters by about 34% relative to the full VGG16 baseline, with public code released for reproducibility. Overall, HCR-Net demonstrates strong cross-script performance and practical potential for script-independent handwriting recognition, while noting areas for future expansion to large-character-set languages and hierarchical architectures.

Abstract

Handwritten character recognition (HCR) remains a challenging pattern recognition problem despite decades of research, and lacks research on script independent recognition techniques. {\color{black}This is mainly because of similar character structures, different handwriting styles, diverse scripts, handcrafted feature extraction techniques, unavailability of data and code, and the development of script-specific deep learning techniques. To address these limitations, we have proposed a script independent deep learning network for HCR research, called HCR-Net, that sets a new research direction for the field. HCR-Net is based on a novel transfer learning approach for HCR, which \textit{partly utilizes} feature extraction layers of a pre-trained network.} Due to transfer learning and image augmentation, HCR-Net provides faster and computationally efficient training, better performance and generalizations, and can work with small datasets. HCR-Net is extensively evaluated on 40 publicly available datasets of Bangla, Punjabi, Hindi, English, Swedish, Urdu, Farsi, Tibetan, Kannada, Malayalam, Telugu, Marathi, Nepali and Arabic languages, and established 26 new benchmark results while performed close to the best results in the rest cases. HCR-Net showed performance improvements up to 11\% against the existing results and achieved a fast convergence rate showing up to 99\% of final performance in the very first epoch. HCR-Net significantly outperformed the state-of-the-art transfer learning techniques and also reduced the number of trainable parameters by 34\% as compared with the corresponding pre-trained network. To facilitate reproducibility and further advancements of HCR research, the complete code is publicly released at \url{https://github.com/jmdvinodjmd/HCR-Net}.

Paper Structure

This paper contains 29 sections, 10 equations, 6 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Organization of rest of the paper:
  3. Literature review
  4. Conventional approaches
  5. Deep learning approaches
  6. HCR-Net
  7. Proposed architecture
  8. Convolution layers
  9. Pooling layers (PLs)
  10. Batch-normalization layers
  11. Dropout layers
  12. Dense layers
  13. Transfer learning
  14. Image augmentation
  15. Two phase training
  16. ...and 14 more sections

Figures (6)

  • Figure 1: Handwritten character recognition research
  • Figure 2: Proposed architecture of HCR-Net
  • Figure 3: An example of mage augmentation using rotation, translation, shear, zoom and hybrid operations
  • Figure 4: Convergence analysis of HCR-Net on Gurmukhi_1.1 dataset, where the first phase takes the first 30 epochs and the second phase uses the remaining 20 epochs.
  • Figure 5: Performance comparison of HCR-Net against state-of-the-art transfer learning techniques for HCR
  • ...and 1 more figures