[ PAPER ID: 37069 ] TEXT DETECTION AND RECOGNITION USING DBNet
ARTICLE INFO: Date of Submission: Oct 30, 2025, Revised: Nov 26, 2025, Accepted: Dec 03, 2025, https://doi.org/10.56815/ijmrr.v4i4.2025.112-123, HOW TO CITE: Suman, Mala M V & Sowmya S (2025). Text Detection and Recognition using Dbnet. International Journal of Multidisciplinary Research & Reviews, 4(4), 112-123.
Abstract
Diverse backgrounds, irregular text shapes, and varying illumination make text detection and recognition in natural scenes a fundamental challenge in computer vision. Recent developments in deep learning have led to more robust solutions, with the Differentiable Binarization Network (DBNet) proving to be a particularly effective framework. DBNet presents a learnable binarization method that adaptively distinguishes text areas from intricate backgrounds, enhancing boundary precision and detection accuracy. In contrast to traditional methods that rely on segmentation, DBNet focuses on optimizing the thresholding mechanism directly. This enables both end-to- end training and efficient inference. Its ability to adapt makes it especially well-suited for managing text instances that are curved, multi-oriented, and small-scale.This study investigates the use of DBNet for detecting and recognizing scene text, underscoring its capacity to produce detailed probability maps and to integrate smoothly with recognition modules. Experimental assessments show that DBNet outperforms competitors across various bench- mark datasets, providing a scalable and dependable solution for practical text comprehension tasks. In automated document analysis and visual information re- trieval, scene text detection plays a crucial role. Methods based on segmentation can often have issues with efficiency and boundary precision, particularly when dealing with complex backgrounds. DBNet tackles these issues by implementing a thresholding mechanism that can be learned, allowing for high-precision detection in real time. DBNet reduces post-processing overhead and achieves faster inference while maintaining precision by adaptively refining probability maps. This study examines the function of DBNet in speeding up text detection pipelines and shows its efficacy across various datasets, underscoring its promise for use in resource-limited contexts like mobile and embedded systems.
