Atom Search Optimization with Deep Learning Enabled Arabic Sign Language Recognition for Speaking and Hearing...

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Sign language is vital for communication among individuals with hearing and speaking impairments.
• Arabic Sign Language (ASL) recognition presents challenges due to complexity and intraclass similarity.
• Bridging the communication gap for the hearing impaired is an ongoing research area.

Purpose of the Study:

• To design an advanced model for Arabic Sign Language recognition to aid communication for disabled individuals.
• To improve the accuracy and efficiency of sign language recognition systems.
• To leverage deep learning and optimization algorithms for enhanced sign language classification.

Main Methods:

• A novel approach combining Atom Search Optimization (ASO) with a Deep Convolutional Autoencoder (DCAE) for sign language recognition (ASODCAE-SLR).
• Utilizing a weighted average filtering for input frame pre-processing.
• Employing a Capsule Network (CapsNet) as a feature extractor and DCAE for sign recognition, with ASO optimizing hyperparameters.

Main Results:

• The ASODCAE-SLR model demonstrated superior performance in Arabic Sign Language recognition tasks.
• Experimental validation using the Arabic Sign Language dataset confirmed the model's effectiveness.
• The ASO algorithm significantly enhanced the efficacy of the DCAE model for recognition.

Conclusions:

• The proposed ASODCAE-SLR model offers a promising solution for improving communication for individuals with hearing and speaking disabilities.
• The integration of ASO and DCAE provides a robust framework for complex sign language recognition.
• Further research can explore broader applications and datasets for sign language recognition systems.