A comprehensive step of XAI for Bengali and English text detection and recognition from natural scene images

A critical first step in many applications, including augmented reality, document analysis, and scene comprehension, is text detection from images. Even though text identification for the English language has advanced significantly, there are particular difficulties when applying these methods to languages such as Bengali because of variations in script, morphology, and typographic characteristics. This article offers a wonderful method for leveraging the Faster R-CNN framework to identify and categorize text from images that contain Bengali and English sentences. The study utilizes a dataset named "Bengali and English text image dataset (BETID)" that consists of images with text in both Bengali and English scripts, taken from a variety of sources such as books, documents, natural scenes, internet images, and landscapes. Pre-processing methods are used to improve the resilience and generalization of the model, including image scaling, normalization, and augmentation. The Faster R-CNN architecture is tuned and trained to identify text regions in images and categorize Bengali or English text. The performance of the text Classification model is evaluated using evaluation measures such as accuracy on the BETID dataset. The proposed Faster R-CNN model achieved an average of 97.90% accuracy, which is better than other deep learning models. Empirical findings showcase the capacity of transfer learning to alleviate data scarcity concerns and enhance text detection precision for low-resource languages such as Bengali.)