Transformer-Based Bidirectional Attention Network for Segmentation-Free Word-Level Text Recognition with Overlapping Characters

Word-level text recognition continues to present a challenging task in document and scene analysis, particularly in the case of overcrowded characters, where conventional methods based on segmentation fail. This paper proposes a non-segmentation approach for text recognition based on transformer architecture with bidirectional attention. The model exploits an encoding-decoding transformer on top of a convolutional visual feature extraction module that permits a global contextual understanding across character sequences. This method does not rely on temporal convolutional networks or recurrent units but otherwise incorporates the bidirectional attention mechanisms added to improve alignment in sequences, character prediction, and so forth, without clear segmentation. We evaluate our model on benchmark datasets of synthetic and real-world overlapping words, which exhibit dramatic improvements over previous CNN-CTC and TCN methods in terms of accuracy and robustness. The architecture sets the limits of state of the art on end-to-end text recognition through complex visual distortions and brings to light the promise of transformer-based solutions in the area of fine-Grain character interaction.

• Publication year

  2025

• Venue

  2025 International Conference on Electronics and Computing, Communication Networking Automation Technologies (ICEC2NT)

• Publication date

  2025-09-03

• Fields of study

  Not labeled

• Identifiers
  DOI 10.1109/ICEC2NT65402.2025.11380006
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition

  2020cited by this paper
• Aggregation Cross-Entropy for Sequence Recognition

  2019cited by this paper
• Decoupled Attention Network for Text Recognition

  2019cited by this paper
• Sequence-To-Sequence Domain Adaptation Network for Robust Text Image Recognition

  2019cited by this paper
• What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis

  2019cited by this paper
• Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition

  2018cited by this paper
• NRTR: A No-Recurrence Sequence-to-Sequence Model for Scene Text Recognition

  2018cited by this paper
• Rosetta: Large Scale System for Text Detection and Recognition in Images

  2018cited by this paper
• Comparison of HMM- and SVM-based stroke classifiers for Gurmukhi script

  2017cited by this paper
• A fuzzy approach to segment touching characters

  2017cited by this paper
• Attention is All you Need

  2017cited by this paper
• Temporal Convolutional Networks for Action Segmentation and Detection

  2016cited by this paper
• Semi-Supervised Learning for Neural Machine Translation

  2016cited by this paper
• An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition

  2015cited by this paper
• A robust method for coarse classifier construction from a large number of basic recognizers for on-line handwritten Chinese/Japanese character recognition

  2014cited by this paper
• Neural Machine Translation by Jointly Learning to Align and Translate

  2014cited by this paper
• A neuro-fuzzy inference engine for Farsi numeral characters recognition

  2010cited by this paper
• Wavelet Adaptive Observer Based Control for a Class of Uncertain Time Delay Nonlinear Systems with Input Constraints

  2009cited by this paper
• Wavelet Adaptive Output Tracking Control for a Class of Delayed Uncertain MIMO Nonlinear Systems Subjected to Actuator Saturation

  2009cited by this paper
• A novel adaptive morphological approach for degraded character image segmentation

  2005cited by this paper
• Novel neural modulators.

  2003cited by this paper
• Segmentation of touching characters in printed Devnagari and Bangla scripts using fuzzy multifactorial analysis

  2001cited by this paper
• On-Line and Off-Line Handwriting Recognition: A Comprehensive Survey

  2000cited by this paper
• A Theory for Multiresolution Signal Decomposition: The Wavelet Representation

  1989cited by this paper

• No citing papers are available for this paper.