Performance Study of LDA and KFA for Gabor Based Ear Recognition System using Distance Metrics

Ear recognition has gained significant attention in recent years due to its potential applications in biometric security systems. Gabor filters are widely used in these systems for their ability to capture the unique texture and shape characteristics of the ear. Unlike many existing studies that rely on pre-processed datasets, this research introduces a custom-collected dataset from diverse subjects, simulating real-world variability in image quality. The images were manually cropped, resized, and normalized for brightness and contrast using Matrix Laboratory (MATLAB) software, ensuring consistency in feature extraction. This study applies Linear Discriminant Analysis (LDA) and Kernel Fisher Analysis (KFA) to the extracted Gabor features, focusing on dimensionality reduction. The results show that KFA outperforms LDA in terms of recognition accuracy and computational efficiency. Additionally, the choice of distance metric is shown to significantly impact performance, offering insights into optimizing biometric systems for real-world applications, especially in diverse populations.