F-norm distance metric based robust 2DPCA and face recognition

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Two-dimensional principal component analysis (2DPCA) is a dimensionality reduction technique.
• The standard 2DPCA uses squared F-norm, which is sensitive to outliers in data.
• Outliers can negatively impact the effectiveness of dimensionality reduction algorithms.

Purpose of the Study:

• To develop a robust 2DPCA algorithm resistant to outliers.
• To improve the characterization of data's geometric structure.
• To enhance the performance of 2DPCA in applications like face recognition.

Main Methods:

• Replaced the squared F-norm with the F-norm as the distance metric in the 2DPCA objective function.
• Developed a novel non-greedy algorithm for optimizing the objective function.
• The proposed algorithm provides a closed-form solution in each iteration, ensuring maximization of the criterion function.

Main Results:

• The F-norm based 2DPCA demonstrates significant robustness against outliers.
• The method effectively captures the underlying geometric structure of the data.
• Experimental results on face databases show superior effectiveness and robustness compared to existing robust 2DPCA algorithms.

Conclusions:

• The proposed F-norm based 2DPCA offers a more robust and effective dimensionality reduction approach.
• This method is particularly beneficial for datasets containing outliers, such as in face recognition.
• The algorithm's ability to preserve geometric data structures enhances its applicability in various machine learning tasks.