Assistive communication system using deep sparse autoencoder with feature learning to assist people with hearing disabilities

Hearing loss is a sensory damage, a hidden disability that affects numerous adults globally. However, gesture recognition plays a significant role in overcoming several problems and challenges in human life, particularly for persons with hearing impairments and speech disabilities. The ability of machines to comprehend human actions and interpret sensory data is exploited across a wide range of applications. Automatic gesture recognition is used to present a response in a computerised learning environment to practice sign language for the deaf community. Deaf people worldwide face challenges conveying their feelings to others. To address this, a smart system powered by artificial intelligence and the Internet of Things was developed to improve communication for the deaf community. The large number of SLs hinders the application’s performance. Current progress in artificial intelligence and deep learning helps automate and enhance gesture recognition systems. This paper proposes an Assistive Communication System Using Deep Sparse Autoencoder with Feature Learning (ACS-DSAEFL) approach to aid the hearing disabled. The ACS-DSAEFL approach aims to develop an effective framework that enhances real-time communication and assistive interactions for people with hearing disabilities. Initially, the image pre-processing stage applies noise removal using a Gaussian filter (GF) and contrast enhancement via histogram equalisation to improve image quality. To achieve effective feature representation, the Inception-enhanced vision transformer (IEViT) model is employed to extract both local and global features. Finally, the stacked sparse denoising autoencoder (SSDAE) is implemented for classification. The comparison assessment of the ACS-DSAEFL technique showed an accuracy of 98.63% compared with other methods on the HaGRID dataset.