Robust Tensor Decomposition for Image Representation Based on Generalized Correntropy.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Traditional tensor decomposition methods like 2D PCA and 2D SVD are susceptible to outliers.
• Minimizing mean square errors in these methods leads to sensitivity issues with noisy data.

Purpose of the Study:

• To propose a novel robust tensor decomposition method resistant to outliers.
• To enhance the performance of tensor decomposition in the presence of noisy data.

Main Methods:

• Developed a new robust tensor decomposition method named Corr-Tensor.
• Utilized the generalized correntropy criterion for objective function optimization.
• Employed a Lagrange multiplier method for iterative optimization.

Main Results:

• Corr-Tensor demonstrates improved robustness against outliers in tensor decomposition.
• The method significantly reduces reconstruction error in face reconstruction tasks.
• Achieved improved accuracies in handwritten digit recognition and facial image clustering.

Conclusions:

• Corr-Tensor offers a robust and computationally efficient alternative to traditional tensor decomposition methods.
• The proposed method effectively handles outliers, leading to better performance in various computer vision applications.