Eigenface and Fisherface Performance Across Pose

资 源 简 介

Eigenface and Fisherface Performance Across Pose

详 情 说 明

Eigenface and Fisherface are two classical methods in face recognition, each with distinct approaches to handling facial variations such as pose changes.

Eigenface relies on Principal Component Analysis (PCA) to reduce the dimensionality of face images by capturing the most significant variations in the dataset. While effective for frontal faces under controlled conditions, its performance degrades when faces are captured from different angles. This is because PCA focuses on global features and may not generalize well to pose variations, leading to recognition errors.

Fisherface, on the other hand, employs Linear Discriminant Analysis (LDA), which not only reduces dimensions but also maximizes the separation between different classes (individuals). This makes Fisherface more robust against pose variations compared to Eigenface, as it emphasizes discriminative features that remain stable across different angles.

In real-world scenarios where faces are rarely perfectly aligned, Fisherface tends to outperform Eigenface due to its ability to preserve meaningful distinctions despite pose changes. However, both methods have limitations under extreme angles, where more advanced techniques like 3D modeling or deep learning may be necessary.

For developers working on face recognition systems, understanding these trade-offs helps in choosing the right algorithm based on the expected pose variability in the application.