BicTd-Brightening Intercultural Communication: Text Detection with Enhanced Light

Identifying text in natural scenes poses significant challenges due to variations in illumination, complex backgrounds, and diverse text styles. This paper introduces novel method aimed at improving the performance over enhancing the detection of low light images with text. Proposed method integrates traditional computer vision techniques with deep learning models to achieve robust and accurate text detection results. Initially, input images are processed to evaluate light values, determining the need for contrast enhancement using adaptive threshold to improve text visibility under different lighting conditions. Following this convolutional neural network architecture is employed to extract features from the video frame. To further refine text detection, proposed work introduce a novel attention mechanism that prioritizes regions with high text likelihood, thereby enhancing the discriminative power of the model. Additionally, a post-processing step based on non-maximum suppression to filter out false positives and improve the final detection results. The success of the implemented technique is demonstrated by an experimental evaluation on an in-house dataset consisting of 20 video clips with multilingual text. It outperforms state-of-the-art algorithms in terms of robustness and detection accuracy in difficult lighting conditions. BicTd method exhibits promising results in real-world scenarios, offering practical solutions for applications such as autonomous driving, document analysis, and augmented reality.