A Robust Tensor-Based Submodule Clustering for Imaging Data Using l12 Regularization and Simultaneous Noise Recovery...

Area of Science:

• Computer Science
• Data Science
• Artificial Intelligence

Background:

• Massive data generation, including images and videos, presents challenges in data management and analysis.
• Real-world data often suffers from noise corruption and lacks labels, necessitating robust unsupervised clustering techniques.
• Traditional image clustering methods that vectorize data fail to preserve essential geometrical structures.

Purpose of the Study:

• To formulate a robust tensor-based submodule clustering method for improved data analysis.
• To enhance clustering capability by incorporating l12 regularization and tensor nuclear norm (TNN).
• To simultaneously address noise removal while preserving data structure in unsupervised clustering.

Main Methods:

• A novel tensor-based submodule clustering method is proposed, utilizing l12 regularization.
• The method incorporates the l12 induced tensor nuclear norm (TNN) to enhance low rankness and self-expressiveness.
• A simultaneous noise removal technique is employed by transforming lateral image slices to frontal slices and applying sparse and low-rank decomposition.

Main Results:

• The proposed method demonstrates superior performance in clustering compared to existing state-of-the-art techniques.
• Experiments were conducted on datasets with sparse, Gaussian, and salt-and-pepper noise, showing effective noise elimination.
• The method successfully retains the geometrical structure of images during the clustering process.

Conclusions:

• The developed tensor-based submodule clustering method offers an effective solution for managing and analyzing large, noisy, and unlabeled image data.
• The integration of l12 regularization and TNN significantly improves clustering accuracy and noise robustness.
• This approach advances unsupervised learning for image data by preserving structural information and removing noise effectively.