Indian Sign Language (ISL) Recognition: An Economical Vision-Based Approach for Hearing & Speech Impaired People

The communication gap between the hearing and speech impaired community and the general population remains a significant social challenge. This paper presents an economical and non-intrusive vision-based system for Indian Sign Language (ISL) recognition, designed to translate static hand gestures into text and speech. The proposed methodology eliminates the need for expensive hardware like data gloves by leveraging a simple webcam for image acquisition. The core of the system involves a robust image processing pipeline: skin color segmentation in the CIEL*a*b* color space, Otsu’s thresholding for binarization, and feature extraction using centroid calculation and the Convex Hull algorithm. These extracted features are classified using a Learning Vector Quantization (LVQ) neural network, chosen for its efficacy in handling complex image datasets with poorly defined cluster boundaries. The system was trained on a dataset of 24 ISL alphabets and several word gestures, achieving accurate real-time recognition. The output is presented both as text and synthesized speech, providing a bidirectional communication aid. This work demonstrates a practical, software-centric solution that enhances accessibility and independence for the deaf and dumb community.