Braille Text Recognition System for the English Alphabet Using Separable CNN and Contour Segmentation of Braille Text

Braille text recognition system is an automated system aimed at enhancing accessibility for visually impaired users. The proposed system combines deep learning models to deliver a robust solution for identifying and segmenting Braille characters, specifically recognizing the 26 English alphabet characters. A Separable Convolutional Neural Network (CNN) is trained to recognize individual Braille characters, utilizing an augmented dataset with noise variations to improve model robustness and accuracy. For segmenting Braille text images into distinct characters, the system employs a contour detection based method, which effectively identifies and outlines individual Braille dots or characters by leveraging grayscale conversion, binary thresholding, and contour analysis. The model achieved a training accuracy of 99.76% with a loss of 0.0300 and a validation accuracy of 99.06% with a loss of 0.0503, and F1 Score with 99.28 % demonstrating its effectiveness and generalization to new data. This work contributes an efficient framework for Braille text recognition, fostering improved accessibility for the visually impaired community.