Principal Component Analysis based on Nuclear norm Minimization.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Principal Component Analysis (PCA) is crucial for dimensionality reduction and feature extraction.
• Traditional PCA is vulnerable to outliers, limiting its real-world application.
• Existing robust PCA methods often neglect error matrix structure and suffer from biased mean estimation.

Purpose of the Study:

• To develop a novel robust Principal Component Analysis (PCA) method.
• To address limitations of existing PCA variants, including outlier sensitivity and structural information neglect.
• To improve the accuracy of low-dimensional data representation and mean estimation in the presence of outliers.

Main Methods:

• Proposed Nuclear Norm-based Principal Component Analysis (N-PCA).
• Developed a unified framework treating data mean, projection matrix, and low-dimensional representation as unknowns.
• Introduced an iterative algorithm with a closed-form solution for each iteration.

Main Results:

• N-PCA effectively utilizes the 2D structure of the error matrix.
• The method mitigates bias in data mean estimation caused by outliers.
• Accurate low-dimensional representation of samples is achieved even with disturbed data.
• Experimental results validate the effectiveness of N-PCA on open databases.

Conclusions:

• N-PCA offers a robust and effective solution for dimensionality reduction in computer vision.
• The proposed unified framework and iterative algorithm provide significant improvements over traditional PCA.
• N-PCA demonstrates superior performance in handling outliers and preserving data structure.