A Comparative Analytical Framework of Decoding Strategies for Offline Handwriting Recognition in Cursive Scripts

Handwritten Text Recognition (HTR) has advanced through deep learning, particularly with decoding architectures such as Connectionist Temporal Classification (CTC), attention-based models, and transducer frameworks. While these approaches achieve high accuracy on Latin scripts, their application to cursive languages such as Arabic, Urdu, Persian, and Pashto remains difficult due to connected letterforms, position-dependent shapes, and diacritic sensitivity. This paper presents a comparative framework that analyzes CTC, attention, and transducer decoding strategies for cursive handwriting recognition. The study reviews their alignment behavior, linguistic modeling capacity, and computational trade-offs. Results indicate that CTC’s monotonic alignment limits robustness to cursive variability, attention-based models improve flexibility but rely on external language models, and transducer-based architecture provides a promising balance through joint alignment and prediction learning. The analysis highlights existing research gaps and offers guidelines for developing decoding methods tailored to the structural and linguistic characteristics of cursive scripts.

• Publication year

  2025

• Venue

  CSI International Symposium on Artificial Intelligence and Signal Processing

• Publication date

  2025-11-22

• Fields of study

  Not labeled

• Identifiers
  DOI 10.1109/AISP68263.2025.11396111
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation

  2025cited by this paper
• HATFormer: Historic Handwritten Arabic Text Recognition with Transformers

  2024influential reference
• OTE: Exploring Accurate Scene Text Recognition Using One Token

  2024cited by this paper
• HTR-VT: Handwritten text recognition with vision transformer

  2024cited by this paper
• Qalam: A Multimodal LLM for Arabic Optical Character and Handwriting Recognition

  2024influential reference
• Finding the reference text in citation contexts using attention model

  2024cited by this paper
• Enhancing Arabic Handwritten Recognition System Based CNN-BLSTM Using Generative Adversarial Networks

  2024cited by this paper
• Data Augmentation for Offline Arabic Handwritten Text Recognition Using Moving Least Squares

  2024cited by this paper
• Handwritten text recognition using deep learning techniques: A survey

  2024cited by this paper
• ViTSTR-Transducer: Cross-Attention-Free Vision Transformer Transducer for Scene Text Recognition

  2023cited by this paper
• Performance Analysis of Vision Transformer Based Architecture for Cursive Handwritten Text Recognition

  2023cited by this paper
• A transformer-based approach for Arabic offline handwritten text recognition

  2023influential reference
• Arabic Handwritten Recognition Using Deep Learning: A Survey

  2022cited by this paper
• An End-to-End OCR Framework for Robust Arabic-Handwriting Recognition using a Novel Transformers-based Model and an Innovative 270 Million-Words Multi-Font Corpus of Classical Arabic with Diacritics

  2022influential reference
• PHTI: Pashto Handwritten Text Imagebase for Deep Learning Applications

  2022influential reference
• A REVIEW ON ARABIC HANDWRITING RECOGNITION

  2022cited by this paper
• Attention-based Deep Learning Model for Arabic Handwritten Text Recognition

  2022influential reference
• CALText: Contextual Attention Localization for Offline Handwritten Text

  2021influential reference
• TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models

  2021cited by this paper
• Recurrent neural network transducer for Japanese and Chinese offline handwritten text recognition

  2021cited by this paper
• OCFormer: A Transformer-Based Model For Arabic Handwritten Text Recognition

  2021influential reference
• Rethinking Text Line Recognition Models

  2021cited by this paper
• Exploring Deep Learning Approaches to Recognize Handwritten Arabic Texts

  2020cited by this paper
• On Recognizing Texts of Arbitrary Shapes with 2D Self-Attention

  2019cited by this paper
• Pioneer dataset and automatic recognition of Urdu handwritten characters using a deep autoencoder and convolutional neural network

  2019influential reference
• NRTR: A No-Recurrence Sequence-to-Sequence Model for Scene Text Recognition

  2018cited by this paper
• Consecutive Convolutional Activations for Scene Character Recognition

  2018cited by this paper
• Sequence Modeling with CTC

  2017cited by this paper
• Attention is All you Need

  2017cited by this paper
• Neural Machine Translation by Jointly Learning to Align and Translate

  2014cited by this paper
• Connectionist Temporal Classification

  2012cited by this paper
• Sequence Transduction with Recurrent Neural Networks

  2012cited by this paper
• Novel neural modulators.

  2003cited by this paper
• Introduction to Transformers

  2001cited by this paper

• No citing papers are available for this paper.