Filipino sign language alphabet recognition using Persistent Homology Classification algorithm.

Area of Science:

• Computer Science
• Artificial Intelligence
• Computational Topology

Background:

• Communication challenges persist for the growing deaf population, with limited support for Filipino Sign Language (FSL).
• Deep networks for FSL recognition face limitations including low transparency, interpretability issues, and high data/computational demands.

Purpose of the Study:

• To explore Topological Data Analysis (TDA) as a novel approach for Filipino Sign Language (FSL) recognition.
• To evaluate the performance of a TDA-inspired classifier, Persistent Homology Classification Algorithm (PHCA), against traditional classifiers for static FSL alphabet recognition.

Main Methods:

• Implemented and evaluated the Persistent Homology Classification Algorithm (PHCA) for classifying static FSL signs.
• Compared PHCA's performance against classical classifiers, including Support Vector Machine (SVM), using balanced and imbalanced datasets with hyperparameter tuning.

Main Results:

• PHCA achieved a mean accuracy of 99.45%, performing comparably to Support Vector Machine (SVM) at 99.31%.
• PHCA demonstrated robustness and stability against data perturbations and noise.
• PHCA exhibited challenges in classifying FSL signs with similar gestures, a common issue in FSL recognition.

Conclusions:

• PHCA presents a viable alternative for Filipino Sign Language (FSL) recognition, offering advantages in transparency and interpretability over deep networks.
• The study highlights the potential of Topological Data Analysis (TDA) in addressing limitations of current machine learning models for sign language recognition.