Abstract
For many individuals, communication through sign language (SL) is the primary means of interacting with the world, and the potential applications of effective SL Recognition (SLR) systems are vast and far-reaching. SLR is a research area dedicated to the automatic analysis of hand gestures and other visual signs used in communication among individuals with speech or hearing impairments. Despite significant advancements, the automated detection and interpretation of human signs remain a complex and multidisciplinary challenge that is yet to be fully addressed. Recently, various approaches have been explored, including the application of machine learning (ML) models in SLR. With advancements in deep learning (DL), sign recognition systems have become more accurate and adaptable, helping to bridge the communication gap for individuals with hearing impairments. Building upon these developments, the present study introduces a novel approach by integrating an advanced optimization strategy with a representation learning model, aiming to improve the robustness, accuracy, and real-world effectiveness of SLR systems. This paper proposes a Pathfinder Algorithm-based Sign Language Recognition System for Assisting Deaf and Dumb People Using a Feature Extraction Model (PASLR-DDFEM) approach. The aim is to enhance SLR techniques to help individuals with hearing challenges communicate effectively with others. Initially, the image pre-processing phase is performed by using the Gaussian filtering (GF) model to improve image quality by removing the noise. Furthermore, the PASLR-DDPFEM approach utilizes the SE-DenseNet model for feature extraction. Moreover, the Elman neural network (ENN) model is implemented for the SLR classification process. Finally, the parameter tuning process is performed by using the Pathfinder Algorithm (PFA) model to enhance the classification performance of the ENN classifier. An extensive set of simulations of the PASLR-DDPFEM method is accomplished under the American SL (ASL) dataset. The comparison study of the PASLR-DDPFEM method revealed a superior accuracy value of 98.80% compared to existing models.