Work

Revolutionizing sign language recognition for hearing-impaired persons using ensemble of deep learning techniques with fine tuning model

Year: 2025

Type: article

Abstract: Sign language has been a major factor in the development of assistive communication devices due to its ability to facilitate interaction between hearing-impaired and hearing people. But still, the biggest hurdle is finding ways of converting gestures into words or sounds with the least possible delay and the highest possible accuracy. Here, we introduce ISRTDO-DLHIP, a state-of-the-art hybrid system for sign language recognition that integrates image preprocessing via adaptive bilateral filtering (ABF), feature extraction using InceptionV3, and a stronger ensemble of BiLSTM, GRU, and VAE classifiers with the Tasmanian Devil Optimization (TDO) algorithm. The presented hybrid method merges metaheuristic optimization with multi-model deep learning to radically enhance recognition reliability and precision. Laboratory tests using the American Sign Language (ASL) dataset have demonstrated that ISRTDO-DLHIP surpasses the existing deep learning models by an average of 98.18% accuracy, + more than + a significant improvement in both precision and recall. This method could be considered a dependable and fast alternative to real-time signing interpretation, thus helping the communication technology to be more inclusive.

Source: Journal of King Saud University - Computer and Information Sciences

DOI: 10.1007/s44443-025-00395-7

Authors: Bayan Alabduallah, Amani A. Alneil

Institutions: Princess Nourah bint Abdulrahman University, Prince Sattam Bin Abdulaziz University, King Salman Center for Disability Research

Language: English

Topic: Hand Gesture Recognition Systems

Subfield: Human-Computer Interaction

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $1,990

Funder: King Salman Center for Disability Research