Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{184229,
        author = {Shilpi Goyal and Dr. Deepak Motwani},
        title = {WTR: Wild Text Recognition Framework through Vision Transformer Dual Attention Mechanism},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {4},
        pages = {4667-4675},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=184229},
        abstract = {Recognition and identification of text in an image is quite difficult due to the prevailing distortions, curved texts, asymmetrical alignments and irregular character spacing. Currently transformer-based methods are used for scene text recognition. It has to place reliance on global self-attention which can recognize horizontally aligned text but unable to identify fine grained characters in complicated wild scenes. This paper explores a new framework called WTR (Wild Text Recognition) which proposes to be a set of dual attention tool which uses global attention for sequence coherence and local attention for fine-grained character features. Two-dimensional learnable positional encodings, within the Vision Transformer encoder, has been used in order to augment spatial adaptability. It further enhances recognition accuracy using language-aware post-processing module which is based on pre-trained language models. Under standard experimental conditions, it has been found that WTR method exceeds other existing state of the art methods as it achieves 96.4% accuracy on IIIT5K, 94.2% on Total-Text, and 96.4% on SCUT-CTW1500, showing robustness against curved, rotated, and perspective-distorted text.},
        keywords = {Scene Text Recognition, Vision Transformer, Dual Attention, 2D Positional Encoding, Language Model.},
        month = {September},
        }