CNN-BiLSTM model for English Handwriting Recognition: Comprehensive Evaluation on the IAM Dataset

TL;DR

This paper introduces a CNN-BiLSTM model for offline English handwriting recognition, achieving state-of-the-art accuracy on the IAM dataset through extensive evaluation, data augmentation, and error analysis.

Contribution

The study presents a comprehensive evaluation of CNN-BiLSTM models for handwriting recognition, including novel test-time augmentation techniques and detailed error analysis.

Findings

Achieved 3.59% CER and 9.44% WER on IAM dataset.

Test-time augmentation reduced WER by 2.5 percentage points.

Provided open-source code to support reproducibility.

Abstract

We present a CNN-BiLSTM system for the problem of offline English handwriting recognition, with extensive evaluations on the public IAM dataset, including the effects of model size, data augmentation and the lexicon. Our best model achieves 3.59\% CER and 9.44\% WER using CNN-BiLSTM network with CTC layer. Test time augmentation with rotation and shear transformations applied to the input image, is proposed to increase recognition of difficult cases and found to reduce the word error rate by 2.5\% points. We also conduct an error analysis of our proposed method on IAM dataset, show hard cases of handwriting images and explore samples with erroneous labels. We provide our source code as public-domain, to foster further research to encourage scientific reproducibility.