Comparative Analysis of Feature Extraction Techniques for Facial Paralysis Classification

Facial paralysis significantly affects a person's ability to communicate and perform essential functions. Facial paralysis classification plays a vital role in the diagnosis and monitoring of facial disorders. Traditional diagnostic methods often rely on subjective evaluations, leading to inconsistent outcomes. The aim of this study is to evaluate and compare various feature extraction techniques to enhance the accuracy and efficiency of facial paralysis classification. The primary contribution of this research lies in its comprehensive analysis of texture-based (Local Binary Patterns, Histogram of Oriented Gradients, Gabor filters) and geometric feature extraction methods, providing insights into their respective strengths and limitations for facial paralysis detection. This study utilizes the YouTube Facial Palsy (YFP) dataset, comprising annotated images of paralyzed and non-paralyzed faces. Preprocessing included resizing images to 128x128 pixels to standardize inputs. Feature extraction methods were applied to the dataset, and the extracted features were classified using machine learning algorithms, including Support Vector Machines (SVM), Random Forest (RF), and k-Nearest Neighbors (KNN). Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The best-performing method achieved an accuracy of 85% using HOG features combined with KNN. The findings highlight that texture-based methods, particularly HOG, excel in capturing subtle asymmetries, while geometric features offer computational efficiency and interpretability with fewer extracted features. This study underscores the importance of selecting suitable feature extraction methods based on task requirements, and emphasizes the potential of hybrid approaches to leverage the strengths of different methods. Future research should explore advanced geometric descriptors and integrate hybrid models to enhance clinical applicability